The Solution of the k(GV) Problem (ICP Advanced Texts in Mathematics)

The Solution of the k (GV ) Problem This page intentionally left blank ICP Advanced Texts in Mathematics – Vol. 4 ...

Author: Peter Schmid

57 downloads 537 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

The Solution of the k (GV ) Problem

This page intentionally left blank

ICP Advanced Texts in Mathematics – Vol. 4

The Solution of the k (GV ) Problem Peter Schmid

Mathematisches Institut der Universität Tübingen, Germany

ICP

Imperial College Press

Published by Imperial College Press 57 Shelton Street Covent Garden London WC2H 9HE Distributed by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data Schmid, Peter, 1941– The solution of the k(GV) problem / by Peter Schmid. p. cm. -- (ICP advanced texts in mathematics ; v. 4) Includes bibliographical references and index. ISBN-13: 978-1-86094-970-8 (hardcover : alk. paper) ISBN-10: 1-86094-970-3 (hardcover : alk. paper) 1. Kernel functions. I. Imperial College of Science, Technology, and Medicine (Great Britain) II. Title. QA353.K47S36 2007 515'.7223--dc22 2007039136

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

Copyright © 2007 by Imperial College Press All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

Printed in Singapore.

EH - The Solution of the k(GV).pmd

1

3/20/2008, 11:00 AM

To the memory of Walter Feit

v

This page intentionally left blank

Preface The k(GV ) problem has recently been solved, completing the work of a series of authors over a period of more than forty years. The objective of this book is to describe the developments, the ideas and methods, leading to this remarkable result. All details of the proof will be presented. Let G be a finite group. The number k(G) of conjugacy classes of G, which is just the number of irreducible complex characters of G, is certainly an invariant of G of special interest. For some families of finite groups, like the symmetric groups and some of the finite classical groups, this invariant is known to some extent. From the point of view of abstract group theory, however, little can be said about k(G) in relation to the group G. Of course k(G) ≤ |G|, with equality if and only if G is abelian. If N is a normal subgroup of G, then it is easy to see that k(G) ≤ k(N ) · k(G/N ) but there are examples where k(N ) > k(G) (e.g. in the dihedral group of order 10). This makes it difficult to bound k(G) by inductive methods. A unifying notion in group theory is the concept of representation. So finite groups often appear as subgroups of permutation groups or linear groups. The “geometry” of a group (as permutation group or linear group, or as a group of Lie type etc.) should be used in order to describe basic invariants. Here we just assume that G is embedded into the linear group GL(V ) of some finite vector space V over some prime field Fp = Z/pZ, and we wish to bound k(G) by a function of |V | = pm . A weak form of the k(GV ) problem is the question whether k(G) ≤ |V | − 1 in the case where G is a p0 -group (p - |G|). Here equality can happen since GL(V ) contains cyclic subgroups of order |V |−1, the so-called Singer cycles. The restriction to p0 -groups is essential, because GL(V ) = GLm (p) has abelian subgroups of order pb

m2 4

c

.

Now consider the semidirect product GV (also written GnV or V : G). Since k(GV ) > k(G), one can strengthen the question above by asking whether k(GV ) ≤ |V | when G is a p0 -subgroup of GL(V ). This is known as the k(GV ) conjecture. The k(GV ) conjecture is a special case of Brauer’s celebrated k(B) problem [Brauer, 1956]. The k(B) conjecture predicts that the number k(B) of ordinary irreducible characters in a p-block B is bounded above vii

viii

The Solution of the k(GV) Problem

by the order of a defect group D of B (Problem (X) in [Feit, 1982]). In the situation of the k(GV ) conjecture, the group GV has a single p-block with the unique defect group D = V . [Brauer and Feit, 1959] established the general upper bound k(B) ≤ 1 + 41 |D|2 . This bound has since only been slightly improved. The conjectured estimate k(B) ≤ |D| has been proved for cyclic and rank 2 abelian defect groups D, and for some families of finite groups, including the symmetric and finite general linear groups. There is reason to believe that the k(B) conjecture is one of the most difficult problems proposed by Brauer. [Nagao, 1962] noticed that the k(B) conjecture holds for p-solvable groups provided it holds for semidirect products as discussed above. So the k(GV ) theorem implies Brauer’s conjecture for such groups. This was the original motivation for treating the k(GV ) problem which is, on the other hand, of interest in its own right. In the proof of the k(GV ) conjecture Clifford theory plays an important role. This theory may be loosely described as using normal subgroups to pass from the module of interest to lower-dimensional, more tractable modules. This can be fruitfully applied to many problems in group representation theory. In a typical sequence of reductions one tries to show that the module in question may be assumed to be irreducible, then absolutely irreducible, then primitive, then tensor indecomposable, and finally tensor primitive. If one tries this approach on the k(GV ) problem, where V is a faithful coprime Fp G-module, the first two steps are easy, but the third is not. Indeed, the final stage of the proof of the k(GV ) conjecture was the rather difficult verification of the problem when V is induced from a certain module with cardinality 52 . Nevertheless, Clifford theory is crucial, in a manner we now wish to explain. We consider the centralizers (stabilizers) CG (v) as v ranges over the vectors in V . It is easy to see that k(GV ) is the sum over the k(CG (v)) when v ranges over a set of representatives for the G-orbits on V . Remarkably, however, the main theoretical results in the solution of the k(GV ) problem assert that the existence of one vector v in V with CG (v) satisfying suitable conditions is enough to imply that k(GV ) ≤ |V |. The most elementary of these centralizer criteria asserts that it suffices to find v ∈ V such that CG (v) = 1 (Theorem 1.5d of this book). Knörr established two more general centralizer criteria which had great influence on later work [Knörr, 1984]. He showed first that k(GV ) ≤ |V | if CG (v)

Preface

ix

is abelian for some v in V (Theorem 3.4d). Even more important was his second criterion, because it allowed one to assume, in many cases, that V is primitive, thus partially overcoming the major obstacle to the direct Clifford-theoretic approach. Knörr’s ideas are related to some general techniques developed in [Brauer, 1968]. Gow advantageously reworked Knörr’s ideas and proved the result in the case that V is a self-dual G-module [Gow, 1993]. In [Robinson, 1995] it was noticed that it suffices to find v ∈ V such that the restriction of V to CG (v) is self-dual. The most powerful criterion then was established in [Robinson and Thompson, 1996]. The Robinson–Thompson criterion (Theorem 5.2b) asserts that k(GV ) ≤ |V | if there exists v in V such that the restriction of V to CG (v) contains a faithful self-dual summand (with real-valued Brauer character). Such a vector v will be called a real vector. This criterion is sufficiently easy to verify, and it is compatible with the Clifford-theoretic reduction, and so led to much progress towards a solution of the k(GV ) problem. At the end of the reduction steps one is left with a pair (G, V ) admitting no real vectors, such that the generalized Fitting subgroup of G has the form E · Z(G), where E is normal in G and absolutely irreducible on V , and where either E is a group of extraspecial type or is quasisimple. Such a pair (G, V ) will be called nonreal reduced. Robinson and Thompson already showed that this can happen only when the characteristic p ≤ 530 . In Chapters 6 and 7, we use counting arguments to show that nonreal reduced pairs do exist only when p is 3, 5, 7, 11, 13, 19 or 31 and V has dimension 2, 3 or 4. It follows that the k(GV ) conjecture is true when the characteristic p is not one of these seven primes. We give a new treatment of the extraspecial case, using ideas from [Robinson, 1997], [Riese–Schmid, 2000], [Gluck–Magaard, 2002a] and [Riese, 2002]. In the quasisimple case our approach is based on work of [Liebeck, 1996], [Goodwin, 2000], [Riese, 2001] and [Köhler–Pahlings, 2001]. Much is simplified by systematically using properties of characters related to extraspecial groups (Heisenberg groups). The counting techniques usually lead to “small” groups G and modules V of “small” degrees, often affording minimal “Weil characters”. After the analysis of the extraspecial and quasisimple cases was completed around 2000, one had to deal with arbitrary pairs (G, V ) admitting no real vectors. In some Clifford-theoretic sense then (G, V ) “involves” a nonreal reduced pair, but a priori this seemed to be hard to control. So it

x


was striking, and a surprise, that V is just obtained by module induction from a nonreal reduced pair, and that G is close to being a full wreath product with respect to the corresponding imprimitivity decomposition of V (Theorems 8.4 and 8.5c). This crucial result was accomplished in [Riese– Schmid, 2003] using ideas developed in [Gluck–Magaard, 2002b]. In the final stages of the proof of the k(GV ) theorem, one therefore had to treat imprimitive modules induced from nonreal reduced modules. Here one requires bounds on the number of conjugacy classes in a permutation group. Liebeck and Pyber have established the general upper bound 2n−1 for the number of conjugacy classes in a permutation group of degree n (Theorem 9.3). In this manner the proof was completed by Riese and Schmid for p 6= 5, and by Gluck, Magaard, Riese and Schmid for p = 5. Chapter 11 addresses the question of when one can have equality k(GV ) = |V |, without giving a conclusive answer, however. In the final two chapters we briefly describe some consequences of the k(GV ) theorem for general block theory and consider the problem of bounding k(GV ) when G is completely reducible on V but |G| and |V | are not coprime. At present the proof of the k(GV ) conjecture relies on the classification of the finite simple groups; see Chapters 7 and 9. But often I argue on the basis of general counting arguments, which do not refer to a certain simple group and which reduce the discussion to groups of low order. Most challenging are indeed the classical groups. For solvable groups the proof is independent of the classification theorem. I hope this monograph will be comprehensible to a graduate student with some background in group theory and representation theory. Much of this general background is provided by Isaacs’ book [Isaacs, 1976]. Some knowledge of the finite simple groups is also needed; the “Atlas of Finite Groups” [Conway et al., 1985] contains much of the necessary information. For the convenience of the reader appendices are included on the cohomology of finite groups, on parabolic subgroups of some finite classical groups, and on the Weil characters of such groups. I am indebted to Walter Feit for drawing my attention to the k(GV ) problem, and for his long-standing support. Thanks also to David Gluck, Kay Magaard and Udo Riese for critical comments and valuable suggestions during the preparation of this book. March 2007

Peter Schmid, T¨ ubingen

Contents

Preface

vii

1. Conjugacy Classes, Characters, and Clifford Theory

1

1.1 Class Functions and Characters

1

1.2 Induced and Tensor-induced Modules

3

1.3 Schur’s Lemma

4

1.4 Brauer’s Permutation Lemma

6

1.5 Algebraic Conjugacy

7

1.6 Coprime Actions

9

1.7 Invariant and Good Conjugacy Classes

10

1.8 Nonstable Clifford Theory

12

1.9 Stable Clifford Theory

13

1.10 Good Conjugacy Classes and Extendible Characters 2. Blocks of Characters and Brauer’s k(B) Problem

18 19

2.1 Modular Decomposition and Brauer Characters

19

2.2 Cartan Invariants and Blocks

21

2.3 Defect and Defect Groups

23

2.4 The Brauer–Feit Theorem

25

2.5 Higher Decomposition Numbers, Subsections

26

2.6 Blocks of p-Solvable Groups

28

2.7 Coprime Fp X-Modules

31

3. The k(GV ) Problem

32

3.1 Preliminaries

32

3.2 Transitive Linear Groups

34

3.3 Subsections and Point Stabilizers

36

3.4 Abelian Point Stabilizers

41 xi

xii


4. Symplectic and Orthogonal Modules

45

4.1 Self-dual Modules

45

4.2 Extraspecial Groups

47

4.3 Holomorphs

49

4.4 Good Conjugacy Classes Once Again

54

4.5 Some Weil Characters

56

4.6 Symplectic and Orthogonal Modules

60

5. Real Vectors

63

5.1 Regular, Abelian and Real Vectors

63

5.2 The Robinson–Thompson Theorem

66

5.3 Search for Real Vectors

68

5.4 Clifford Reduction

71

5.5 Reduced Pairs

74

5.6 Counting Methods

74

5.7 Two Examples

77

6. Reduced Pairs of Extraspecial Type

82

6.1 Nonreal Reduced Pairs

82

6.2 Fixed Point Ratios

84

6.3 Point Stabilizers of Exponent 2

86

6.4 Characteristic 2

90

6.5 Extraspecial 3-Groups

92

6.6 Extraspecial 2-Groups of Small Order

96

6.7 The Remaining Cases 7. Reduced Pairs of Quasisimple Type

103 110

7.1 Nonreal Reduced Pairs

110

7.2 Regular Orbits

112

7.3 Covering Numbers, Projective Marks

115

7.4 Sporadic Groups

119

7.5 Alternating Groups

121

7.6 Linear Groups

125

Contents

xiii

7.7 Symplectic Groups

129

7.8 Unitary Groups

136

7.9 Orthogonal Groups

145

7.10 Exceptional Groups

147

8. Modules without Real Vectors

148

8.1 Some Fixed Point Ratios

148

8.2 Tensor Induction of Reduced Pairs

149

8.3 Tensor Products of Reduced Pairs

155

8.4 The Riese–Schmid Theorem

156

8.5 Nonreal Induced Pairs, Wreath Products

160

9. Class Numbers of Permutation Groups

170

9.1 The Partition Function

170

9.2 Preparatory Results

171

9.3 The Liebeck–Pyber Theorem

172

9.4 Improvements

174

10. The Final Stages of the Proof

180

10.1 Class Numbers for Nonreal Reduced Pairs

180

10.2 Counting Invariant Conjugacy Classes

182

10.3 Nonreal Induced Pairs

185

10.4 Characteristic 5

186

10.5 Summary

194

11. Possibilities for k(GV ) = |V |

195

11.1 Preliminaries

195

11.2 Some Congruences

197

11.3 Reduced Pairs

199

12. Some Consequences for Block Theory

202

12.1 Brauer Correspondence

202

12.2 Clifford Theory of Blocks

203

12.3 Blocks with Normal Defect Groups

207

xiv


13. The Non-Coprime Situation

209

Appendix A: Cohomology of Finite Groups

213

Appendix B: Some Parabolic Subgroups

217

Appendix C: Weil Characters

221

Bibliography

225

List of Symbols

230

Index

231

Chapter 1

Conjugacy Classes, Characters, and Cliord Theory For the proof of the k(GV ) theorem many of the standard methods and techniques from ordinary and modular representation theory will be applied. In this section we describe the necessary concepts and tools from ordinary character theory. The reader is referred to [Isaacs, 1976] for the relevant background and some basic results used but not proved here. This book is cited as [I] in the text. 1.1.

Class Functions and Characters

Fix a finite group X of order |X| and exponent exp (X) = e. Let K = Q(ε) for ε = e2πi/e . Each complex character χ of X has its values in K, even in its ring of integers Z[ε], because for each x ∈ X there is a basis such that an underlying representation carries x to a diagonal matrix consisting of eth roots of unity, and χ(x) is the trace of this matrix. This χ is constant on conjugacy classes of X, a class function, and if c = cx is the conjugacy class of X containing x (cx = xX = {xt = t−1 xt| t ∈ X}), we may write P χ(c) = χ(x). The distinct class sums b c = y∈c y form a basis of the centre Z(CX) of the semisimple group algebra CX (and of Z(KX)). It follows that the set C`(X) of conjugacy classes of X is in bijection with the ordinary (complex) irreducible characters of X, i.e., (1.1a)

k(X) = |C`(X)| = |Irr(X)|.

If χ, ψ are K-valued class functions on X, their inner product is denoted P P 1 1 by hχ, ψi = hχ, ψiX = |X| x∈X χ(x)ψ(x) = |X| c∈C`(X) |c| · χ(c)ψ(c). One knows that Irr(X) is an orthonormal basis for the K-vector space of class functions on X [I, 2.17]. The (nonsingular) square matrix X = (χ(c)) χ∈Irr(X) (rows and columns somehow arranged) is the character table c∈C`(X)

of X. For some simple groups (of small order) these tables can be found in [Conway et al., 1985], which we usually refer to as the [Atlas]. 1

2


Orthonormality of the irreducible characters may be expressed by the t matrix equation X · T · X = I, where T is the diagonal matrix with entries 1 |CX (x)| for x ∈ c and CX (x) is the centralizer (|c| = |X : CX (x)|). This t

gives X · X = T−1 and hence the second orthogonality relations:

(1.1b)

P

½ χ∈Irr(X) χ(x)χ(y) =

|CX (x)| if xX = y X . 0 otherwise

P 2 In particular |X| = χ∈Irr(X) χ(1) is the sum over the squares of the degrees of the irreducible characters. A generalized character χ of X is a rational integer combination of irreducible characters of X (χ ∈ Z[Irr(X)]). A subgroup E of X is called p-elementary if E = P × Z where P is a p-group for some prime p and Z is a cyclic p0 -group, and it is elementary if it is p-elementary for some prime p. We have Brauer’s characterization of characters: Theorem 1.1c (Brauer). Let χ be a complex valued class function on X. Then χ ∈ Z[Irr(X)] if and only if the restriction ResX E (χ) ∈ Z[Irr(E)] for any elementary subgroup E of X. In fact, each generalized character of X is a Z-linear combination of characters of X induced from linear characters of elementary subgroups. For a proof see [I, 8.10 and 8.12]. Induced characters will be discussed in Sec. 1.2 below. Since linear characters (of degree 1) can be realized over their value fields, using elementary properties of the Schur index we get: Theorem 1.1d (Brauer). K is a splitting field for X, that is, for all characters χ of X there is a matrix representation over K or, equivalently, a KX-module affording χ. It is immediate, then, that K is a splitting field for all subgroups of X. The Schur index will be briefly discussed in Sec. 1.3. If V and W are (right) KX-modules affording the characters χ, ψ, then V ⊕ W affords χ + ψ (sum), and the KX-module V ⊗K W (diagonal X-action) affords χψ (product). V ⊗K V = Sym2 (V )⊕Alt2 (V ) decomposes into the symmetric squares and alternating squares.

Conjugacy Classes, Characters, and Clifford Theory

3

1.2. Induced and Tensor-induced Modules Suppose Y is a subgroup of X and W is a (right) KY -module affording the character θ. Let n = |X : Y | be the index of Y in X, and let Y \X = {Y t | t ∈ X} denote the (transitive) X-set with respect to right multipliT cation (Y t, x) 7→ Y tx. Then the normal core N = CoreX (Y ) = t∈X Y t is the kernel of this permutation representation. Hence G = X/N is a transitive subgroup of the symmetric group Sn . If Y 6= X (n > 1), then S t t∈X Y 6= X and there are at least n − 1 permutations in G without fixed points. See also [Serre, 2003] for a recent discussion of this classical result by Jordan. Theorem 1.2a (Frobenius). There is an embedding of X into the wreath product Y wr Sn = Y(n) : Sn , which is uniquely determined up to conjugacy. The wreath product is defined by letting Sn act on the nth direct power Y (n) of Y permuting the direct factors (so (yi )πi = (yi )iπ = (yiπ−1 )i , sending an entry in the ith position to the iπth position). The wreath product may be identified with the group of all monomial n × n-matrices with entries in Y . “The” Frobenius embedding is obtained by choosing a right transversal {ti }ni=1 to Y in X. Associate then to x ∈ X the element (xi ) · πx in the wreath product, where πx : i 7→ ix in G ⊆ Sn and xi ∈ Y are defined by ti x = xi tix . Replacing {ti } by {yi ti } for certain yi ∈ Y leads to an embedding conjugate under (yi )−1 . Now the base group B = Y (n) of the wreath product acts diagonally onto W (n) = W ⊕ · · · ⊕ W (n direct summands) via (wi ) · (yi ) = (wi yi ), and Sn through (wi )i · π = (wiπ−1 )i . This makes W (n) into a K[Y wr Sn ]module. The induced KX-module V = IndX Y (W ) is obtained through “the” Frobenius embedding of X into Y wr Sn (where conjugate embeddings yield isomorphic module structures). Fixing a transversal {ti } the character of X afforded by V is given by (1.2b)

IndX Y (θ)(x) =

Pn i=1

θ(ti xt−1 i ),

where we set θ(·) = 0 for elements outside Y . Induction and restriction are maps on class functions adjoint to each other, namely related by the Frobenius reciprocity (1.2c)

X hIndX Y (θ), χiX = hθ, ResY (χ)iY .

4


In terms of modules this says that if W is a KY -module, V a KX-module, every KY -homomorphism f : W → ResX Y (V ) extends uniquely to a KXX X homomorphism IndY (f ) : IndY (W ) → V . We often will use Mackey decomposition for characters (and modules). Suppose Y and H are subgroups of X. Then H acts on X\Y , and if {rj } is a set of representatives for the distinct H-orbits (or double cosets of X mod (Y, H)), for a character θ of Y we have (1.2d)

X ResX H (IndY (θ)) =

P j

rj

Y rj IndH Y rj ∩H (ResY rj ∩H (θ )).

Here for x ∈ X we define the (conjugate) character θx of Y x = x−1 Y x by θx (y x ) = θ(y) for y ∈ Y . One can define IndX Y (W ) = W ⊗KY KX (viewing KX as a left KY module). Let now W ⊗n = W ⊗ · · · ⊗ W (n factors, the tensors over K). Again B = Y (n) acts diagonally on W (n) via (⊗i wi ) · (yi ) = ⊗i wi yi , and Sn acts as (⊗i wi ) · π = ⊗i wiπ−1 . This makes W ⊗n into a K[Y wr Sn ]-module. ⊗n Then the tensor-induced KX-module Vb = TenX is obtained Y (W ) = W through “the” Frobenius embedding of X into the wreath product. Mackey decomposition (for modules) carries over to tensor induction in the obvious (multiplicative) way. So if {rj } is a set of representatives for the cycles (orbits) of an element x ∈ X in its action on Y \X, and if the jth cycle has size nj (so that rj xnj rj−1 ∈ Y ), then (1.2e)

TenX Y (θ)(x) =

Q j

θ(tj xnj t−1 j )

describes the character χ b of X afforded by Vb = TenX Y (W ). In order to prove this it suffices to consider the case where Vb = W ⊗ W x ⊗ · · · ⊗ W xn−1 (with xn ∈ Y ). If b is a basis of W consisting of eigenvectors for xn , then the wi0 ⊗ wi1 x ⊗ · · · ⊗ win−1 xn−1 , with wij ∈ b, form a basis of Vb for which the matrix of x is monomial. Only basis vectors of the form vi = wi ⊗ wi x ⊗ · · · · ⊗wi xn−1 , with fixed wi ∈ b, are eigenvectors of x on Vb , and if wi xn = ci wi with ci ∈ K, then vi x = ci vi . Hence χ b(x) = θ(xn ), as desired. 1.3. Schur’s Lemma Suppose V and W are KX-modules affording the characters χ, ψ, respectively. Then hχ, ψi = dim K HomKX (V, W).


5

∼K If χ, ψ are irreducible, HomKX (V, W ) = 0 if χ 6= ψ and EndKX (V ) = if V = W (Schur’s lemma). Of course this yields the first orthogonality relations. In particular, if ρ : X → GLn (K) is a (matrix) representation of X affording χ ∈ Irr(X), only the scalar matrices are permutable with all ρ(x), x ∈ X. Extending ρ linearly to KX we get a K-algebra homomorphism into Mn (K) and, by restriction to the centre, a homomorphism ωχ : Z(KX) → K, the central character associated to χ. If c = cx = xX is the conjugacy class of x ∈ X, (1.3a)

ωχ (b c) =

χ(x)|X: CX (x)| χ(1)

=

χ(c)|c| χ(1) .

The values of ωχ are algebraic integers since the product of class sums is a (nonnegative) integer linear combination of class sums. This is important P c)χ(c) for block theory (Chapter 2). Writing 1 = hχ, χi = χ(1) c∈C`(X) ωχ (b |G| |G| is an integer (being rational and an algebraic integer). we see that χ(1) Clifford theory will even yield the following (see also [I, 3.12 and 6.15]).

Theorem 1.3b (Itô). The degree χ(1) of an irreducible character χ of X divides |X : V | for any abelian normal subgroup V of X. We used that irreducible representations over K are absolutely irreducible. Replacing K by the field K0 = Q(χ) generated by the values of χ, there is a unique irreducible K0 X-module V , up to isomorphism, whose character contains χ. Then V affords the character mχ where m = m(χ) is the Schur index of χ (over the rationals). D = EndK0 X (V ) is a K0 -division algebra with centre K0 and dimension m2 [I, 9.21]. We give some further examples where Schur’s lemma is involved. Let x ∈ X and χ ∈ Irr(X), and let χ ¯ be the complex conjugate character. Then −1 χ(x) ¯ = χ(x) = χ(x ) (see Sec. 1.5). We assert that P (1.3c) |χ(x)|2 = (χχ)(x) ¯ = χ(1) y∈X χ([x, y]). |X| Here [x, y] = x−1 xy denotes the commutator of x and y. Let ρ be a representation of X affording χ. The sum on the right is the trace of P y −1 χ(1) cx )|CX (x)| = ρ(x−1 )χ(x), and the ρ(x−1 ) χ(1) ) |X| ωχ (b y∈X ρ(x ) = ρ(x |X| assertion follows. We see that |χ(x)| = χ(1) if and only if χ([x, y]) = χ(1) for all y ∈ X. Note that Ker(χ) = {x ∈ X| χ(x) = χ(1)} is the kernel of ρ, and Z(χ) = {x ∈ X| |χ(x)| = χ(1)} consists of those x ∈ X for which ρ(x) is a scalar matrix.

6


Suppose next that X = hY, xi for some subgroup Y , and that θ = ResX Y (χ) is (still) irreducible. We assert that then (1.3d)

P y∈Y

|χ(xy)|2 = |Y |.

The 1-character 1X is contained in χχ ¯ with multiplicity 1, because we have ¯ = 1. Clearly h1X , χχi ¯ = hχ, χi = 1. Similarly h1Y , ResX ¯ = h1Y , θθi Y (χχ)i X ResY (1X ) = 1Y . So if ψ 6= 1X is an irreducible constituent of χχ, ¯ ResX Y (ψ) does not contain 1Y . Hence if W is a KX-module affording ψ and τ is an P underlying representation, then y∈Y τ (y) = 0 as dim K CW (Y) = dim K HomKY (K, W) = h1Y , ψiY = 0. P P It follows that y∈Y ψ(xy) = 0 by considering the trace of τ (x) y∈Y τ (y). P P We conclude that y∈Y (χχ)(xy) ¯ = y∈Y 1X (xy) = |Y |. 1.4.

Brauer’s Permutation Lemma

Suppose G is a finite group acting on the finite set Ω (from the right, by permutations). Then we write C`(G|Ω) = orb(G on Ω) for the set of orbits of G on Ω. So C`(X) = C`(X|X) with X acting by conjugation. By the Cauchy–Frobenius fixed point formula (1.4a)

|C`(G|Ω)| =

1 |G|

P g∈G

|CΩ (g)|.

This is sometimes also called Burnside’s lemma; it is easily proved by means of the counting principle or using Frobenius reciprocity [Serre, 2003]. Each orbit is a (transitive) G-set and so isomorphic to H\G for some subgroup H, which is determined up to conjugacy in G (being a point stabilizer). The isomorphism type of the G-set Ω is determined by the “marks” |CΩ (H)| for all (nonconjugate) subgroups H of G [I, 13.23]. We associate to the G-set Ω the permutation character πΩ of G, counting the fixed points of each element. Theorem 1.4b (Brauer). Suppose G is a finite group which acts on Irr(X) and on C`(X) such that χg (cg ) = χ(c) for all χ ∈ Irr(X), c ∈ C`(X) and g ∈ G. Then for each g ∈ G, the number kg (X) = |CC`(X) (g)| of g-invariant conjugacy classes agreas with the number |CIrr(X) (g)| of ginvariant irreducible characters of X. In particular, the permutation characters πC`(X) = πIrr(X) of G agree.


7

The proof [I, 6.32] is based on the fact that the character matrix of X is nonsingular. It follows, in view of Eq. (1.4a), that G has the same number of orbits on Irr(X) and on C`(X). Of course, this does not mean that these sets are permutation isomorphic (unless G is cyclic). 1.5. Algebraic Conjugacy Let Γ = Gal(K|Q) be the Galois group of K over the rationals. We have a natural action of Γ on Irr(X). We also have a permutation action on X as Γ∼ = (Z/eZ)? , where xσ = xn if σ ∈ Γ corresponds to the coset of n modulo e = exp (X). This preserves C`(X). Let χ ∈ Irr(X). If εi ∈ K are the eigenvalues of x appearing in a representation to χ, the eigenvalues for xn are the εni . It follows that χσ (x) = χ(x)σ = χ(xn ) = χ(xσ ). In order to apply Theorem 1.4b one has to alter the action of Γ on X (say) −1 by assigning (x, σ) 7→ xσ . This works since Γ is abelian. Notice that if σ is complex conjugation (restricted to K), χσ (x) = χ(x) ¯ = χ(x) = χ(x−1 ). So the number of real-valued irreducible characters of X is equal to the number of real conjugacy classes c of X, satisfying c−1 = c (Burnside). X is called a real group if all its conjugacy classes are real, that is, if all χ ∈ Irr(X) are real-valued. Suppose X has odd order. Then {1} is the unique real class of X. Since P k(X)−1 P χ(1) = χ(1) ¯ is odd, we get |X| = χ∈Irr(X) χ(1)2 = 1+2 i=12 (1+2ni )2 for certain integers ni ≥ 1. Consequently (1.5a)

|X| ≡ k(X) (mod 16),

a well known result due to Burnside. Recall integers. If αi over the α is a root have (1.5b)

that a character of X takes only values which are algebraic α 6= 0 is such an algebraic integer with the distinct conjugates Pn rationals (1 ≤ i ≤ n), then i=1 |αi | ≥ n, with equality only if of unity. For by the arithmetic–geometric mean inequality we

1 n

P

|αi | ≥ (

Q

1

1

|αi |) n = |N(α)| n ,

8


Q with equality only if all |αi | are equal. But the norm N(α) = i αi is a nonzero rational integer. Hence the assertion holds, with equality only if all |αi | are equal and N(α) = ±1. In this case |αi | = 1 for all i, whence α is a root of unity (since only finitely many powers of α are distinct). Lemma 1.5c (Gallagher). Suppose y ∈ X is such that χ(y) 6= 0 for each χ ∈ Irr(X). Let N = [hyi, X]. Then k(X) ≤ |CX (y)| − (|X/N | − k(X/N )). Proof. We follow [Gallagher, 1962]. N is the (normal) subgroup of X generated by all commutators [t, x], t ∈ hyi, x ∈ X. By the second orthogonality relations (1.1b), |CX (y)| =

X

|χ(y)2 | = Σ1 + Σ2 ,

χ∈Irr(X)

where the first sum is over those χ with |χ(y)| = χ(1) and the second sum is over the others. Now |χ(y)| = χ(1) if and only if y ∈ Z(χ) (as described in Sec. 1.3), and this happens if and only if N is in the kernel of χ and so χ may be viewed as a character of X/N . Thus Σ1 = |X/N |, and the number of irreducible characters of X in Σ1 is equal to k(X/N ) = |Irr(G/N )|. For each σ in Γ = Gal(K|Q) we have |χ2 |σ = (χ · χ) ¯ σ = χσ χσ = |χσ |2 . Thus Σ1 and Σ2 are Galois stable. By hypothesis the average over the Galois class of |χ(y)2 | is ≥ 1 (and is equal to 1 only if χ(y) is a root of unity). Consequently Σ2 ≥ k(X) − k(X/N ), and the result follows. ¤ Theorem 1.5d. Suppose X has an abelian normal Sylow p-subgroup, V , for some prime p. Then X = GV for some p-complement G in X, uniquely determined up to conjugacy. For each v ∈ V , ¡ ¢ k(GV ) ≤ |CG (v)| · |V | − |G| − k(G) . In particular, if CG (v) = 1 for some v ∈ V , then k(GV ) ≤ |V | and equality only holds if G is abelian. Proof. By a simple cohomological argument X = GV is as claimed (Appendix A1; Schur–Zassenhaus theorem). Let v ∈ V . We assert that χ(v) 6= 0 for each χ ∈ Irr(X). By Theorem 1.3b, χ(1) is not divisible by p. Letting p be a prime ideal above p in the ring of integers of K, we have χ(v) ≡ χ(1) (mod p) (cf. Chapter 2). Hence the assertion. Now N = [hvi, X] = [v, G] is a normal subgroup of X contained in V , and


9

CX (v) = CG (v)V . By an elementary counting argument, carried out in Theorem 1.7a below, k(X/N ) ≤ |V¡/N | · k(G).¢ Thus by the preceding lemma k(X) ≤ |CG (v)| · |V | − |V /N | |G| − k(G) . From CG (v) = 1 it follows that k(X) ≤ |V |, and then k(X) = |V | only if |G| = k(G), that is, if G is abelian. ¤ 1.6.

Coprime Actions

If G is a finite group and V is a finite G-module of order prime to |G|, then all (Tate) cohomology groups Hn (G, V ) vanish (A1). For n = 1, 2 this leads to the Schur–Zassenhaus theorem (already mentioned). For n = 0, −1 this tells us that the fixed module CV (G) is the image of the trace map P v 7→ g∈G vg on V and that the commutator module [V, G] = [V, G, G] is its kernel. Then CV (G) ∩ [V, G] = 0 and so (1.6a)

V = CV (G) ⊕ [V, G].

Irr(V ) = Hom(V, C? ) is the character group of V , and |CIrr(V ) (G)| = |Irr(V /[V, G])| = |V /[V, G]| = |CV (G)|. The corresponding holds for all subgroups of G. Hence we have the following. Proposition 1.6b. If V is a G-module where V and G have coprime order, then V and Irr(V ) are isomorphic G-sets. The proposition is true without assuming that V is abelian. In fact, if G acts on X by automorphisms and |G| and |X| are coprime, then Irr(X) and C`(X) are isomorphic G-sets. This result is due to Isaacs and Dade. Its proof makes use of the Feit–Thompson theorem. So either G is solvable, in which case a proof can be found in [I, 13.24], or X is solvable, where a proof can be found in [Isaacs, 1973]. We do not need this result in this general form. Theorem 1.6c (Glauberman). Suppose G is a cyclic group acting on X by automorphisms where |G| and |X| are coprime. Let ξ be a character of the semidirect product GX = X : G for which χ = ResX (ξ) is irreducible. Then there is a unique irreducible constituent θ of the restriction to Y = CX (G) of χ, a unique linear character µ of G and a unique sign ± such that ξ(gy) = ±µ(g)θ(y) for all generators g of G and all y ∈ Y . If G is a p-group for some prime p, the sign is such that hχ, θiY ≡ ±1 (mod p).

10


This is a special case of a more general character correspondence. For a proof we refer to [I, 13.6 and 13.14]. The character ξb = ξ · µ ¯ is the socalled canonical extension to GX of χ, determined by the fact that its b has X in its kernel. determinantal character det(ξ) 1.7. Invariant and Good Conjugacy Classes Let N be a normal subgroup of X, and let G = X/N . Then G acts on C`(N ) in the natural way, and kg (N ) = |CC`(N ) (g)| is the number of ginvariant conjugacy classes of N for each g ∈ G. The conjugacy class g G of g is called “good for N ” provided CX (x)N/N = CG (g) for any (some) x ∈ g (with N x = g). This is well-defined. Suppose N is abelian. Then the class of g is good for N if CX (x)/N = CG (g) for x ∈ g, that is, whenever a commutator [x, y] ∈ N for some y ∈ X then [x, y] = 1. In this case each conjugacy class of G is good for N if and only if N is central in X and no nontrivial element of N is a commutator in X. Theorem 1.7a (Gallagher). Let Y be a subgroup of X, and let N = CoreX (Y ) be its normal core. Let G = X/N . (i) k(Y ) ≤ |X : Y | · k(X) and k(X) ≤ |X : Y | · k(Y ), the latter inequality being proper unless Y = N . Moreover, (1.7b)

k(X) ≤ k(N ) · k(G),

where equality holds if and only if each conjugacy class of G is good for N . (ii) Let g = N x for some x ∈ X. The conjugacy class in C`(N ) of an element y ∈ N is fixed by g if and only if Cg (y) 6= ∅, and then |Cg (y)| = |CN (y)|. The number of g-invariant conjugacy classes of N is (1.7c)

kg (N ) =

1 |N |

P y∈N

|CN (xy)|.

P 1 Proof. We follow [Gallagher, 1970]. By (1.4a) k(X) = |X| x∈X |CX (x)|. The first inequality in (i) is immediate from CY (x) ⊆ CX (x). Using the inequality |CX (x)| ≤ |X : Y | · |CY (x)| and the counting principle we have X x∈X

|CX (x)| ≤ |X : Y |

X x∈X

|CY (x)| = |X : Y |

X y∈Y

|CX (y)|,


11

P and this is at most equal to |X : Y |2 y∈Y |CY (y)|. We have equality if and only if CX (x)Y = X for all x ∈ X, and in this case any two conjugate elements of X are Y -conjugate. Then Y is normal in X. Before proving (1.7b) we settle (ii). Let t ∈ N . Then tx ∈ Cg (y) ⇐⇒ −1 y txy = tx ⇐⇒ xyx−1 = t−1 yt ⇐⇒ y x = y t . Hence Cg (y) 6= ∅ if and only if y N is fixed by g −1 (or g), and then Cg (y) = CN (y)tx for some t ∈ N . So |Cg (y)| = |CN (y)| if y N ∈ CC`(N ) (g) and Cg (y) = ∅ otherwise. We conclude that X 1 X |Cg (y)|. kg (N ) = 1/|N : CN (y)| = |N | N −1

y∈N

y∈N :y ∈CC`(N ) (g)

Counting the pairs (tx, y) ∈ g × N satisfying (tx)y = y(tx) we see that P P t∈N |CN (tx)| = y∈N |Cg (y)|. This proves Eq. (1.7c), and completes the proof of (ii). For each x ∈ X we have CX (x)/CN (x) ∼ = CX (x)N/N ⊆ CG (N x). Hence X

|CX (x)| ≤

x∈X

and

P

t∈g

X

|CG (N x)| · |CN (x)| =

x∈X

X g∈G

|CG (g)|

X

|CN (t)|,

t∈g

P P |CN (t)| = y∈N |Cg (y)| ≤ y∈N |CN (y)| by (ii). Hence X X X |CX (x)| ≤ |CG (g)| |CN (y)|, x∈X

g∈G

y∈N

where equality holds if and only if CX (N x) = CX (x)N for each x ∈ X (and each X-class of N is an N -class). We are done. ¤ Theorem 1.7d (Keller). Let Y be a proper subgroup of X, N = CoreX (Y ) and G = X/N . Let Ω = C`(N |X) with N acting by conjugation, which is a G-set (with G acting by conjugation). Then we have a partition Ω = U g∈G Ωg where Ωg is the set of N -orbits contained in the coset g. (i) For each g ∈ G the centralizer CG (g) is the stabilizer in G of Ωg , and |Ωg | = kg (N ). (ii) Let g1 = 1, g2 , · · ·, gr , gr+1 , · · ·, gk(G) be representatives for the distinct conjugacy classes of G, the first r classes being just those meeting H = Y /N . Then k(G)

k(X) =

X

|C`(CG (gi )|Ωgi )| ≤ k(Y ) + (k(G) − r) · M

i=1

where M = max{kg (N )| g 6∈

S t∈G

Ht }.

12


Proof. This is a recent result due to [Keller, 2006]. Let g ∈ G. It is obvious that CG (g) is the stabilizer in G of Ωg . By the Cauchy–Frobenius formula (1.4a) and part (ii) of the preceding theorem, X 1 X |Cg (y)| = 1/|N : CN (y)| = kg (N ). |Ωg | = |N | N y∈N

y∈N :y ∈CΩ1 (g)

This proves (i). Each G-orbit on Ω is of the form (xN )G for some unique conjugacy class xX ∈ C`(X), and determines the conjugacy g G of G defined by N x = g or, equivalently, by xN ⊆ g (xN ∈ Ωg ). In particular, k(X) = |C`(G|Ω)|. For h ∈ G we have |C`(CG (g h )|Ωgh )| = |C`(CG (g)|Ωg )|. This yields the identity given in (ii). For 1 ≤ i ≤ r we may pick the representatives gi ∈ H, belonging then to certain distinct conjugacy classes of H. Of course r ≤ k(H), and k(G) > r by Jordan’s theorem. By what is already proved (applied to Y ), r X i=1

|C`(CG (gi )|Ωgi )| ≤

r X

|C`(CH (gi )|Ωgi )| ≤ k(Y ).

i=1

For the remaining k(G) − r conjugacy classes g G of G, for which g is not in S t t∈G H , we take the trivial estimate |C`(CG (g)|Ωg )| ≤ |Ωg |, and use the fact that |Ωg | = kg (N ). ¤ 1.8. Nonstable Clifford Theory Let N be a normal subgroup of X. Then X acts on N via conjugation (as a group of automorphisms), and on Irr(N ). We have induced actions of G = X/N on C`(N ) and on Irr(N ), and Theorem 1.4b applies. Fix θ ∈ Irr(N ). The stabilizer of θ (in X) is called the inertia group T = IX (θ), and T /N = IG (θ). If χ ∈ Irr(X|θ) is an irreducible character of X above θ, that is, θ is a constituent of ResX N (χ), there are just s = |X : T | distinct X-conjugates θ = θ1 , · · ·, θs of θ and Ps (1.8a) ResX N (χ) = eχ i=1 θi for some integer eχ ≥ 1, the ramification index of χ with respect to N . Theorem 1.8b (Clifford). Let T = IX (θ). The map ψ 7→ χ = IndX T (ψ) is a bijection from Irr(T |θ) onto Irr(X|θ). The ramification indices eχ = eψ are divisors of |T /N |. For a proof we refer to [I, 6.11 and 11.29].


13

1.9. Stable Clifford Theory Let again N be a normal subgroup of X, and let G = X/N . Suppose that θ ∈ Irr(N ) is G-invariant (IG (θ) = G). This is a necessary condition for the existence of a character χ of X extending θ. If such a χ exists then Irr(X|θ) = {χλ = χ ⊗ λ| λ ∈ Irr(G)}, and this has cardinality k(G) [I, 6.17]. Moreover, then {χλ| λ ∈ Irr(G), λ(1) = 1} is the set of all characters of X extending θ, and this has cardinality |G/G0 |. Here G0 = [G, G] is the commutator subgroup of G, the kernel of the characters of G of degree 1. In general we proceed as follows. Let K0 = Q(θ) be the field generated by the values of θ, and let W be an irreducible K0 N -module affording mθ, where m = m(θ) is the Schur index. Then D = EndK0 N (W ) is a centrally simple K0 -algebra with dim K0 D = m2 (1.3). Since N is normal in X and θ is stable under G, each conjugate module W g = W ⊗ g (affording θg ) is a K0 N -module isomorphic to W (g ∈ G). Choose ¡ X K0 N¢-isomorphisms L τg : W g → W (with τ1 = idW ). We have ResX N IndN (W ) = g∈G W g, and by Frobenius reciprocity (1.2c) the τ extend uniquely to units in the ¡ X ¢ Lg −1 G-graded ring EndK0 X IndN (W ) = g∈G Dτg . Then τg Dτg = D for all g ∈ G. By the Skolem–Noether theorem [Bourbaki, 1958, Chap. 8, §10] we may choose the τg such that they centralize D (via conjugation). Then −1 τgh τg τh = τ (g, h) · idW for some nonzero scalar τ (g, h) ∈ K0 . We have a projective representation of G with 2-cocycle τ ∈ Z 2 (G, K0? ), where the multiplicative group K0? = (K0 r {0}, ·) is viewed as a trivial Gmodule. The cohomology class of τ depends only on θ, K0 and the group extension N ½ X ³ G; it is written µK0 G (θ). This “Clifford obstruction” is functorial in that it maps onto the corresponding cohomology class when replacing K0 by an extension field. Proposition 1.9a. The Clifford obstruction µK0 G (θ) ∈ H2 (G, K0? ) vanishes if and only if there is a character χ of X extending θ and satisfying K0 (χ) = K0 . The order of µK0 G (θ) is a divisor of the number of |G|th roots of unity in K0 . There is a distinguished central group extension Z ½ G(θ) ³ G, where Z is a cyclic group of order exp (N ), whose cohomology class maps onto µKG (θ) through an (appropriate) embedding of Z into K ? . The exponent of G(θ) is a divisor of e = exp (X). c the strucProof. If µK0 G (θ) vanishes, by definition one can give W = W c c ture of a K0 X-module satisfying EndK0 X (W ) = D. This W affords mχ where χ ∈ Irr(X) extends θ, with m(χ) = m = m(θ). It follows that

14


K0 = Q(χ). The converse is proved similarly. The order of µK0 G (θ) divides |G| by an elementary property of cohomology groups (Appendix A1). It also divides the number of roots of unity in K0 [Dade, 1974] which, however, will not be used here. We briefly discuss the further (basic) statements. Replacing K0 by the complex number field we are just concerned with Schur’s theory of lifting projective representations. The Schur multiplier M(G) = H2 (G, Z) of G fits into the natural universal coefficient exact sequence (Appendix A5) 0 → Ext(G/G0 , K0? ) → H2 (G, K0? ) → Hom(M(G), K0? ) → 0. Passing to the complex number field, and noting that C? is divisible, we see that H2 (G, C? ) is nothing but the dual of M(G). So there is a (complex) character χ of X extending θ if M(G) = 1. Of course, then Q(χ) ⊆ K and so µKG (θ) vanishes. Let K1 = Q(ε1 ) where ε1 is a primitive exp (N )th root of unity, and let Z = hε1 i. By Theorem 1.1d this K1 is a splitting field −1 for N . Let W be a K1 N -module affording θ, and let τ (g, h) = τgh τg τh be a 2-cocycle with class µK1 G (θ). We wish to show that there is a unique element in H2 (G, Z) mapping onto this cohomology class. Consider the long exact cohomology sequence to Z ½ K1? ³ K1? /Z: δ

H1 (G, K1? /Z) → H2 (G, Z) → H2 (G, K1? ) → H2 (G, K1? /Z). Either K1? /Z is torsion-free or |Z| is odd and (−1)Z is its unique torsion element. At any rate, δ is the zero map, and it suffices to show that µK1 G (θ) has trivial image in H2 (G, K1? /Z). In order to prove this, as well as for the proof of the final statement, we may assume that G is a p-group for some prime p (A4). The construction of G(θ) will show that its exponent divides exp (X). Arguing by induction on |X| we may also assume that θ is faithful, because Ker(θ) is normal in X, and that there is no proper subgroup X0 of X covering G such that N0 = X0 ∩N has a G ∼ = X0 /N0 -invariant irreducible character θ0 satisfying µK1 G (θ0 ) = µK1 G (θ) in H2 (G, K1? ). This reduction will lead us, in the case where G is a p-group, to X = G(θ). On the basis of Theorem 1.1c, we find a p-elementary subgroup X0 of X covering G and an X0 -invariant irreducible character θ0 of N0 = X0 ∩N such that hθ0 , θiN0 is not divisible by p [I, 8.24]. But this implies that µK1 G (θ0 ) = µK1 G (θ). In order to see this, let U be a K1 N0 -module affording θ0 , and consider the K1 -space H = HomK1 N0 (U, W ). By Frobenius reciprocity


15

X 0 each f ∈ H extends uniquely to a K1 X0 -morphism IndX ), N0 (U )¡→ IndN (W ¢ X0 preserving the G-gradings. Since the G-graded ring EndK1 X0 IndN0 (U ) = L −1 g∈G K1 σg is a crosssed product, the maps f 7→ σg f τg may be considered as elements ψg ∈ GL(H). We obtain that −1 ψgh ψg ψh = σ(g, h)−1 τ (g, h) · idH −1 where σ(g, h) = σgh σg σh ∈ K1? is a factor set with class µK1 G (θ0 ). Taking determinants we get that σ d and τ d agree modulo the coboundary obtained from g 7→ det(ψg ). Here d = dim K1 H = hθ0 , θiN0 . Since G is a p-group, hence so is H2 (G, K1? ), and since p does not divide d, the cohomology classes of σ and τ agree. Thus by our choice X = X0 is p-elementary.

Let next M be a G-invariant abelian subgroup of N of maximal order. Since X is p-elementary and X/N a p-group, it follows that CN (M ) = M . Let λ ∈ Irr(M ) be an irreducible (linear) constituent of ResN M (θ) and X1 = IX (λ), N1 = IN (λ) = X0 ∩ N . Let θ1 ∈ Irr(N1 |λ) be the unique character satisfying IndN N1 (θ1 ) = θ (Theorem 1.8b). By a Frattini argument X1 covers G and, by the same argument as before, µK1 G (θ1 ) = µK1 G (θ) since hθ1 , θiN1 = 1. Thus X = X1 and N1 = N . It follows that ResN M (θ) = θ(1)λ and that M ⊆ Z(N ) as θ is faithful. But CN (M ) = M . Hence N = M is central in X and θ = λ is linear. Now µK1 G (θ) is the image of the cohomology class of the central extension N ½ X ³ G under the map induced by θ¯ = θ−1 : N → K1? . Indeed, choose τg : w ⊗ tg 7→ w for some transversal {tg }g∈G to N in X. Letting t(g, h) = t−1 gh tg th be the corresponding factor set, µK1 G (θ) is the class of the factor set (g, h) 7→ θ(t(g, h)−1 ) of G, which has its values in Z. ¤ Definition 1.9b. The group G(θ) in the preceding proposition is called the representation group of θ (with respect to G). The extended representation group is defined as the “fibre-product” (pull-back; “diagonal group” in the terminology of the Atlas) X(θ) = G(θ)∆G X −1 of G(θ) and X amalgamating G = X/N . Letting τ (g, h) = τgh τg τh be a 2-cocycle with values in Z (viewed as a group of scalar multiplications) and class µKG (θ) in H2 (G, K ? ), we may write G(θ) as the group consisting of all pairs (g, z) ∈ G×Z with multiplication (g, z)(h, z 0 ) = (gh, zz 0 τ (g, h)). Then X(θ) consists of all elements ((g, z), x) for which N x = g. By Proposition

16


1.9a, X(θ) has the same exponent as X. Hence K is a splitting field for X(θ) by Theorem 1.1d. c gets Suppose W is a KN -module affording θ. By construction W = W the structure of a KX(θ)-module through w((g, z), x) = (zwx)τg = (zw)τg x for w ∈ W ⊆ IndX N (W ), g ∈ G, z ∈ Z and x ∈ X (with N x = g). For c is an extension of W when viewed x ∈ N we have v((1, 1), x) = vx. Thus W ∼ as a module for Ker(X(θ) ³ G(θ)) = N . We may replace K by any subfield which is a splitting field for θ. In this manner we find a character θb of X(θ) extending θ, when viewed as a character of Ker(X(θ) ³ G(θ)), which can be written in this same field. This applies in particular when the Schur index m(θ) = 1 (which is true in prime characteristic). Theorem 1.9c (Clifford). Let N be a normal subgroup of X, let G = X/N and let θ ∈ Irr(N ) be stable in X. Let θb ∈ Irr(X(θ)) extend θ in the above sense, and let θe−1 be the unique irreducible (linear) constituent of θb on Ker(X(θ) ³ X) ∼ = Z. Then ζ ↔ χ = θb⊗ ζ is a 1-1 correspondence between e and Irr(X|θ). Moreover |Irr(X|θ)| = |Irr(G(θ)|θ)| e ≤ k(G). Irr(G(θ)|θ) Proof. Let χ ∈ Irr(X|θ), and view χ as a character of X(θ) by inflation. Since θb extends θ when viewed as a character of Ker(X(θ) ³ G(θ)) ∼ = N, there is a unique (irreducible) character ζ of G(θ) such that χ = θb ⊗ ζ (see above). But Ker(X(θ) ³ X) ∼ = Z is in the kernel of χ. It follows that e e e ζ ∈ Irr(G(θ)|θ) where θ is as described. Conversely, every ζ ∈ Irr(G(θ)|θ) b gives rise to an irreducible character χ = θ ⊗ ζ in Irr(X|θ). For the final statement we may assume that X = G(θ), N = Z is central in X and θe = θ. By Frobenius reciprocity (1.2c) we then have P IndX N (θ) = χ∈Irr(X|θ) χ(1)χ, and this vanishes outside N and agrees with P |G|θ on N . For χ ∈ Irr(X|θ) we have |χ(x)|2 = χ(1) y∈X χ([x, y]) by |X| P (1.3c) and, of course, |X| = x∈X |χ(x)|2 . Hence |Irr(X|θ)| =

1 X |X|

X

|χ(x)|2 =

x∈X χ∈Irr(X|θ)

This is at most equal to k(G).

|G| |X|2

P x∈X

|G| X θ([x, y]). |X|2 x,y∈X [x,y]∈N

P y∈CX (N x)

1 =

1 |G|

P x ¯∈G

|CG (¯ x)| = ¤


17

We now shall discuss Clifford theory of tensor induction. Suppose N is a nonabelian normal subgroup of X which is the central product of the X-conjugates of some (proper) subgroup N0 . Let X0 = NX (N0 ) be the normalizer, and let G = X/N and G0 = X0 /N0 . Assume θ ∈ Irr(N ) is stable in X, and let θ0 ∈ Irr(N0 ) be the unique irreducible constituent of θ on N0 . As θ0 is absolutely irreducible, θ is the (tensor) product of the |X : X0 | distinct X-conjugates of θ0 [I, 4.21]. b = Define X(θ) and X0 (θ0 ) as before, with the same Z, and let Z

IndX X0 (Z) be the (induced) permutation module.

Theorem 1.9d. Keeping these assumptions, let θb0 be an irreducible character of X0 (θ0 ) extending θ0 (as above). Then there is a group extension b (θb ) is a character b ³ X mapping onto X(θ) such that θb = TenX Zb ½ X b0 0 X b b of X(θ) extending θ, X0 being the inverse image in X of X0 . Proof. Let ρ0 : N0 → GL(W0 ) be a K-representation affording θ0 (TheoQ rem 1.1d). Let {ti }ni=1 be a right transversal to X0 in X. Let h = i htii be an element in N (with all hi ∈ N0 ). Then ρ(h) = ⊗ni=1 ρ0 (htii ) is a Nn K-representation of N on W = i=1 W0 ti affording θ. Since θ0 is stable in X0 , we may extend ρ0 to a projective representation ρb0 : X0 → GL(W0 ). We may choose ρb0 such that its factor set τ0 ∈ Z 2 (X0 , K ? ), being inflated from G0 , has order dividing |Z| = exp (N0 ). Let x ∈ X, and let ti x = xi tix be as in Sec. 1.2 (with xi ∈ X0 ). Then ρb(x) = ⊗nix=1 ρb0 (xi ) defines a projective representation of X tensor induced from ρb0 which extends ρ and has factor set n Y τb0 (x, y) = τ0 (xi , yix ). i=1

This τb0 is co-induced from τ0 (and inflated from G). Q xi tix and so i (hi )

We have hx =

ρ(hx ) = ⊗nix=1 ρ0 (hxi i ) = ρb(x)−1 ρ(h)b ρ(x). Thus the class of τ = τb0 in H2 (G, K ? ) is nothing but µKG (θ), and ρb lifts b to an ordinary representation of X(θ), say affording θ. b represents the cohomology class obtained from The group extension X b underlying X0 (θ0 ) under the natural isomorphism H2 (X0 , Z) ∼ = H2 (X, Z) b under the Shapiro’s lemma (A3). Here the group X(θ) is the image of X Qn n b map Z → Z sending (zi ) to zi (zi ∈ Z). ¤ i=1

i=1

18


1.10. Good Conjugacy Classes and Extendible Characters Let N be a normal subgroup of X, and let G = X/N . Let θ ∈ Irr(N ) be G-invariant. The conjugacy class of an element g = N x in G is called “good for θ” provided θ can be extended to hN, x, yi for all y ∈ X satisfying [x, y] ∈ N [Gallagher, 1970]. By virtue of Theorem 1.9c this may be studied by passing to G(θ). Hence we may assume that N = Z is cyclic and central in X and that θ = θe is linear. Assume also that θ is faithful. If there is a (linear) character λ of Y = hN, x, yi extending θ, then Y /N is abelian (as [x, y] ∈ N ), Y /Ker(λ) is abelian and N ∩ Ker(λ) = Ker(θ) = 1. Hence Y is abelian. Conversely, if Y is abelian, then θ can be extended to Y . We have proved the following. Theorem 1.10a. Suppose θ ∈ Irr(N ) is stable under G = X/N . Then |Irr(X|θ)| is the number of conjugacy classes of G which are good for θ. Combining this with Theorem 1.8b we obtain the Clifford–Gallagher formula (1.10b)

k(X) =

P θ∈Irr(N )

¡ ¢ kθ IG (θ) /|G : IG (θ)|,

where kθ (IG (θ)) is the number of conjugacy classes of IG (θ) which are good for θ. Observe that kθ (IG (θ)) = k(IG (θ)) whenever θ can be extended to IX (θ), that is, whenever µKIG (θ) (θ) vanishes. This happens for instance if all Sylow subgroups of IG (θ) are cyclic, because then its Schur multiplier is trivial by (A4).

Chapter 2

Blocks of Characters and Brauer's k(B) Problem Keeping the assumptions and notation of the preceding chapter we proceed now to modular representations and to block theory. The reader is referred to [Feit, 1982] for the relevant background as well as for results not proved here. This book is cited as [F] in the text. We describe some ideas of Brauer and Feit when attacking the k(B) problem. The study of (major) subsections will motivate us (once again) to consider point stabilizers when dealing with the k(GV ) problem. On the basis of the Clifford theory of blocks developed by Fong we shall verify the k(B) problem for p-blocks of p-solvable groups, assuming that the k(GV ) theorem is already settled. 2.1. Modular Decomposition and Brauer Characters Let p be a rational prime. Denote by |X|p = pa the p-part of the order of X (= order of a Sylow p-subgroup of X), and let Xp 0 be the set of p 0 elements (or p-regular elements) of X. Let Z(p) denote the localization of the integers at the prime (p) = pZ. Then R = Z(p) [ε] is a Dedekind ring with only finitely many nonzero prime ideals, all lying above p. Hence R is a principal ideal domain, with quotient field K = Q(ε). Letting p be any of the (Galois conjugate) nonzero prime ideals of R, the field F = R/p is finite of characteristic p, and is a splitting field for all subgroups of X (as follows from Theorem 1.1d, or from the fact that Schur indices are trivial). If χ, ζ are K-valued class functions on Xp 0 , we define hχ, ζip 0 = hχ, ζiXp0 =

1 X χ(x)ζ(x). |X| x∈Xp 0

For irreducible characters χ, ζ of X we define mχζ = hχ, ζip 0 (which is symmetric), and we say that they belong to the same p-block, B, of X provided their central characters agree mod p. See Eq. (1.3a); recall that the central characters have their values in Z[ε] ⊆ R. This defines a partition of Irr(X) (which turns out to be independent of the choice of the maximal 19

20


ideal p). For the time being, the block B is just a certain nonempty set Irr(B) of irreducible characters of X. Define ωB (b c) = ωχ (b c) + p for any χ ∈ Irr(B), and all conjugacy classes c of X. If χ is a class function on X, we let χ e=χ e(pa ) be the class function a on X taking the value p χ(x) for x ∈ Xp0 , and zero otherwise. For χ, ζ in e Irr(X) we have pa mχζ = he χ, ζi = hχ, ζi. Lemma 2.1a. If χ ∈ Z[Irr(X)] is a generalized character, then so is χ e. 1 χ e ∈ Z[Irr(X)] where d ≤ a is the smallest integer (if any) Then even pa−d such that

pd |CX (x)|p χ(x)

∈ R for all x ∈ Xp0 .

Proof. We apply Theorem 1.1c. Let E = P ×Q be an elementary subgroup of X, where P is a p-group and Q is a p0 -group. Then |P | = pb with b ≤ a. Let ρ be the character of P afforded by the regular representation. Then ResX χ) = (pa−b ρ) ⊗ ResX e is a E (e Q (χ) is a generalized character of E. Thus χ 1 χ e ∈ Z[Irr(X)] if and only if generalized character of X. It follows that pa−d h

1 pa−d

χ e, ζi ∈ Z(p) = R ∩ Q

for all ζ ∈ Irr(X). But for x ∈ Xp0 and c = xX we have |c| = |X : CX (x)| and ζ(x−1 ) 1 X 1 pd χ e(y)ζ(y −1 ) = χ(x) · ∈ R. a−d |X| y∈c p |CX (x)| p |CX (x)|p0 e vanishes on p-singular elements of X. Use finally the fact that χ

¤

Reduction mod p is an isomorphism from the exp (X)p 0 th roots of unity in K to those in F . Let V be an F X-module. Each p-regular x ∈ Xp 0 is semisimple (diagonalizable) on V , its eigenvalues being such roots of unity in F . Lifting these eigenvalues, and summing up, we get a class function ϕV : Xp 0 → K. This ϕV is called the Brauer character of X afforded by V (with respect to p). If X is a p 0 -group, ϕV is an ordinary character of X. If V is an irreducible F G-module, ϕV is called an (absolutely) irreducible Brauer character. The set IBr(X) = IBrp (X) of (absolutely) irreducible Brauer characters of X is a basis for the K-vector space of class functions on Xp 0 [I, 15.11], or [F, IV.3.4]. It follows that

Blocks of Characters and Brauer’s k(B) Problem

(2.1b)

21

|IBr(X)| = |C`(Xp 0 )|

is the number of p 0 -classes in X. The reader is referred to [Jansen et al., 1995] for tables of Brauer characters for some simple groups of small orders; this book will be quoted as [B-Atlas]. Theorem 2.1c. Let χ ∈ Irr(X). Then χp 0 = ResX Xp0 (χ) is a Brauer charP acter, hence χp 0 = ϕ∈IBr(X) dχϕ ϕ for some unique nonnegative integers dχϕ , called the decomposition numbers of χ. Proof. By Theorem 1.1d, there is a KX-module W affording χ. Let {wj } be a K-basis of W , and let U be the R-submodule of W generated by all wj x, x ∈ X. Then U is finitely generated and torsion-free, hence a free R-module. U must have rank χ(1) = dim K W, and U is stable under X. Thus U is an RX-lattice affording χ. It follows that V = U/pU is an F X-module affording the Brauer character χp 0 = ResX Xp0 (χ). This χp 0 is independent of the choice of the lattice U affording χ. ¤ 2.2. Cartan Invariants and Blocks Let B be a block of X. We say that an irreducible Brauer character ϕ of X belongs to B, ϕ ∈ IBr(B), provided dχϕ 6= 0 for some χ ∈ Irr(B). Theorem 2.2a. Let B be a block of X, and let ϕ ∈ IBr(B). Then dχϕ = 0 whenever χ does not belong to B, and ϕ is a Z-linear combination of the χp 0 for χ ∈ Irr(B). Moreover: P (i) The associated “projective character” ϕ b = χ∈Irr(B) dχϕ χ vanishes off p-regular elements, its degree ϕ(1) b is divisible by pa , and it satisfies hϕ, b ϕip0 = 1 and hϕ, b ψip0 = 0 for ϕ 6= ψ in IBr(X). b be the “Cartan invariant”, and (ii) For ϕ, ψ in IBr(B) let cϕψ = hϕ, b ψi let CB = (cϕψ )ϕ,ψ . The Cartan matrix CB of B is symmetric and positive −1 definite, and CB = (hϕ, ψip0 )ϕ,ψ . Proof. Suppose dχϕ 6= 0. Let V = U/pU be an F X-module affording χp0 . Then some composition factor of V affords ϕ. Each class sum b c acts as a scalar multiplication on V , thus defining a central character ωϕ (with values in F ). This ωϕ must be the reduction mod p of ωχ . In other words, ωϕ = ωB for some unique p-block B.

22


For the second statement, observe that modular decomposition may be seen as a map from generalized characters to generalized Brauer characters, which commutes with restriction to subgroups and induction (defined for Brauer characters like for ordinary characters). Hence by Brauer’s induction theorem it suffices to consider the case where X = P × Q is elementary (P a p-group, Q a p0 -group). But in this case every ϕ ∈ IBr(X) has P in its kernel, because CV (P ) 6= 0 for each irreducible F X-module V , and CV (P ) is X-invariant as P is normal in X. It follows that ϕ = χp0 for some unique χ ∈ Irr(X). The surjectivity of the decomposition map is equivalent to the statement that the elementary divisors of the decomposition matrix DB = (dχϕ ) of B, with χ ∈ Irr(B) and ϕ ∈ IBr(B), are all 1. For ϕ, ψ in IBr(B) the Cartan invariant X b = cϕψ = hϕ, b ψi dχϕ dχψ χ∈Irr(B) t

and so CB = (DB ) · DB is a positive definite symmetric matrix with nonnegative integer entries. By the second orthogonality relations (1.1b), for any regular x ∈ Xp0 and any singular y ∈ X we have X X b ϕ(x)ϕ(y) χ(x)χ(y) = 0. = ϕ∈IBr(X)

χ∈Irr(X)

Since the irreducible Brauer characters are linearly independent, ϕ(y) b =0 ϕ b (1) is for each ϕ. Thus if P is a Sylow p-subgroup of X, then hϕ, b 1 i = P P

|P |

an integer. Varying c = cx over the p-regular conjugacy classes of X the t second orthogonality relations show that (ϕ(c))ϕ,c · (ϕ(c)) b ϕ,c is the diagonal matrix with cx th entry |CX (x)|. It follows that for ϕ, ψ in IBr(B), X 1 ϕ(x)ψ(x) b hϕ, b ψip0 = = δϕψ . |CX (x)| c=cx ∈C`(Xp0 )

bp0 = From ϕ

P

ψ∈IBr(B) cϕψ ψ

−1 we infer that CB is as asserted.

¤

We define a graph with Irr(X) as its vertex set by linking χ, ζ in Irr(X) if there exists ϕ ∈ IBr(X) such that dχϕ 6= 0 6= dζϕ . This is called the Brauer graph (mod p). By Theorem 2.2a, Irr(B) is a union of P connected components of the Brauer graph. We also let eB = χ∈Irr(B) eχ P where eχ = χ(1) x∈X χ(x)x is the centrally primitive idempotent of KX |X| associated to χ (ωχ (eχ ) = 1 and ωχ (eζ ) = 0 for ζ 6= χ).


23

P Theorem 2.2b (Osima). We have eB = c∈C`(X) aB (c)b c for unique elements aB (c) in R, and we write eB and aB (c) also for the reductions mod p . Then aB (c) = 0 (in F ) if c is a p-singular class, and ωB (eB ) = 1. Also, Irr(B) is a connected component of the Brauer graph. P 1 b = 0 if c Proof. By Theorem 2.2a we have aB (c) = |X| ϕ∈IBr(B) ϕ(1)ϕ(c) is not a p0 -class. If c is a p0 -class, then aB (c) =

1 |X|

X

ϕ(1)ϕ(c) b

ϕ∈IBr(B)

is in R, because pa is divisor of ϕ(1) b by Theorem 2.2a. Replacing Irr(B) by a connected component B ⊆ Irr(B) of the Brauer graph, and IBr(B) by {ϕ ∈ IBr(X)| dχϕ 6= 0 for some χ ∈ B}, the corresponding statement P holds for e = χ∈B eχ . Then an irreducible character χ of X belongs to B if and only if ωχ (e) 6≡ 0 (mod p). We conclude that B = Irr(B). ¤ 2.3. Defect and Defect Groups Let χ ∈ Irr(B). By Theorem 1.3b, χ(1)p ≤ pa . If χ(1)p = pa−d is as small as possible in Irr(B), then d = d(B) is the defect of B, and χ is said to be of height zero in B. In general χ(1)p = pa−d+h with height h = hχ ≥ 0. Theorem 2.3a. Let B be a block of X with defect d and let χ ∈ Irr(B) be of height zero. Then for any ζ ∈ Irr(B), pd mχζ is a nonzero rational −1 integer whose p-part is equal to the height of ζ. In particular pd CB is a positive definite symmetric matrix with integer entries. Proof. Let χ e ∈ Z[Irr(X)] be as in Lemma 2.1a. From Theorem 2.2a it P χ, ζiζ is in follows that mχζ = 0 if ζ ∈ 6 Irr(B). Hence χ e = ζ∈Irr(B) he |G| 1 · χ(1) ∈ R for c = xX and x ∈ Xp0 , pa−d χ e Z[Irr(B)]. Since ωχ (b c) = |Cχ(x) X (x)| 1 e is not in is a generalized character (in B) by Lemma 2.1a. But pa−d+1 χ Z[Irr(B)], because if P is a Sylow p-subgroup of X then

1 pa−d+1

he χ, 1P iP = χ(1)/pa−d+1

is not an integer. It follows that there is θ ∈ Irr(B) such that he χ, θi/pa−d a−d is a unit in R. Then θ(1)p = p . Let ζ ∈ Irr(B) and ζ(1)p = pa−d+h . e e There are similar statements for ζ. By definition pa mχζ = he χ, ζi = hχ, ζi.

24


Thus pd mχζ is an integer whose p-part equals ζ(1)p . Note that hχ,θi ˜ θ(1)

hχ,ζi ˜ θ(1)

≡

(mod p), because ωζ and ωθ agree mod p. For the final statement observe that for χ, ζ ∈ Irr(B) we have mχζ =

X

dχϕ · hϕ, ψip0 · dζψ .

ϕ,ψ∈IBr(B) −1 t Thus (pd mχζ )χ,ζ = DB (pd CB )DB . Now use the fact that by Theorem 2.2a all the elementary divisors of DB are equal to 1. ¤

Let c = xX be a conjugacy class of X. Then a Sylow p-subgroup of CX (x) is a defect group for c. By Theorem 2.2b there exist c such that aB (c) 6= 0 and ωB (b c) 6= 0. Then c is called a defect class for the block B. Lemma 2.3b. Let B be a block of X and let D be a defect group of some defect class for B. Let c be any conjugacy class of X. (a) If aB (c) 6= 0 (in F ), then D contains a defect group for c. (b) If ωB (b c) 6= 0, then D is contained in a defect group for c. For a proof of this Min-Max lemma see [I, 15.31] . It follows that the defect groups of the defect classes for B form a single conjugacy class of p-subgroups of X, called the defect groups of B. If D is a defect group of B then |D| = pd where d is the defect of B. Indeed, let χ ∈ Irr(B) be of height zero, whence χ(1)p = pa−d . Let c = cx be a defect class for B. Then ωχ (b c) 6≡ 0 (mod p) by definition. Since χ(x) ∈ R, this shows that |c|/χ(1) ∈ R ∩ Q = Z(p) is not divisible by p. Hence |D| ≥ pd as |c|p = pa /|D|. On the other hand, also aB (c) 6= 0 and so in view of Theorems 2.2b and 2.2a there is some ζ ∈ Irr(B) such that ζ(c) 6≡ 0 (mod p). Using that ωζ¯(b c) ∈ R it follows that ζ(1)p ≤ |c|p . Hence |D| ≤ pd . The principal block of X is the block containing the 1-character 1X . Its defect groups are the Sylow p-subgroups of X. Theorem 2.3c. Let B be a block of X with defect group D, and let P be a normal p-subgroup of X. Then |D| = pd where d is the defect of B, and D ⊇ P . The block idempotent eB of B is a central idempotent of F CX (P ), and if CX (P ) ⊆ P , then B is the unique p-block of X and hence D a Sylow p-subgroup of X.


25

Proof. We have already proved the first statement. Let N = CX (P ), which is normal in X. Let c ∈ C`(X). If some x ∈ c is contained in N , then c ⊆ N . Assume c 6⊆ N . Then each P -orbit on c (by conjugation) has size divisible by p. Let V be an irreducible F X-module. Then P ⊆ CX (V ) (as seen above). It follows that V b c = 0. Consequently b c is in the Jacobson radical of Z(F X), hence is nilpotent. Now let c be a defect class for B with defect group D. Then ωB (b c) 6= 0 and so b c is not nilpotent. Thus c ⊆ N = CX (P ) and P ⊆ D. P Let eB = c aB (c)b c be as in Theorem 2.2b. We have seen that either b c ∈ J(Z(F X)) or c ∈ C`(X) is centralized by P . Since eB is an idempotent, it follows that eB ∈ Z(F X) ∩ F N . Let b be a block of N covered by B, that is, some χ ∈ Irr(B) lies over some θ ∈ Irr(b). It follows from Theorem 1.8b that the blocks of N covered by B form a unique X-conjugacy class of blocks. Thus eB · eb 6= 0 and eB is P the sum of the distinct X-conjugates of eb . Define ωbX (b c) = ωb ( x∈N ∩c x) for any conjugacy class c of X. If c 6⊆ N , then c ∩ N = ∅ and ωbX (c) = 0 by definition. Then also ωB (b c) = 0, for otherwise D ⊆ CX (x) for some x ∈ c by the preceding lemma, whence x ∈ CX (D) ⊆ CX (P ) = N , a contradiction. If c ⊆ N , then b c ∈ Z(F N ) and ωB (b c) = ωb (b c) = ωbX (b c), because central characters agree if they agree on central idempotents (knowing that F is a splitting field for Z(F N )). We conclude that ωB = ωbX is determined by b, i.e., B = bX is the unique block of X covering b. In particular, if CX (P ) ⊆ P , then the principal block b is the unique p-block of N = Z(P ) and so B is the unique p-block of X. ¤ 2.4. The Brauer–Feit Theorem In what follows we fix a block B of X with defect group D, |D| = pd . We let k(B) = |Irr(B)| and `(B) = |IBr(B)|. We have k(B) ≥ `(B) since the Cartan matrix CB is nonsingular. One even has k(B) > `(B) unless d = 0, in which case k(B) = 1 = `(B) [F, IV.4.19]. Brauer’s celebrated k(B) conjecture is the assertion that always k(B) ≤ pd . Theorem 2.4 (Brauer–Feit). We have k(B) ≤ 1 + 41 p2d . If B contains an irreducible character of positive height, then even k(B) ≤ 21 p2d−2 . Proof. Let kh = kh (B) be the number of irreducible characters in B of P height h, so that k(B) = h≥0 kh . Let χ ∈ Irr(B) be of height zero. By

26


P 1 χ e = ζ∈Irr(B) nζ ζ where each nζ = pd mχζ = Theorem 2.3a, pa−d is a nonzero integer with p-part equal to phζ . From pd nχ = pd (pd mχχ ) = h we get pd nχ ≥ n2χ + (k0 − 1) + k(B) ≤

X h≥0

1 pa−d

P h≥1

χ e,

1 pa−d

χ ei =

X

1 he χ, ζi pa−d

n2ζ

ζ∈Irr(B)

kh p2h . It follows that

1 kh p2h ≤ 1 + pd nχ − n2χ ≤ 1 + p2d , 4

because t 7→ pd t − t2 takes its maximum in t = P P 2h ≤ 41 p2d and so h≥1 kh ≤ 41 p2d−2 . h≥1 kh p

1 d 2p .

We also see that

Suppose there is a character ζ ∈ Irr(B) with height h = hζ > 0. Then ζe ∈ Z[Irr(B)]. Arguing as before this yields that k0 p2 ≤ pd u − u2 for u = pd−h mζζ . Hence k0 ≤ 41 p2d−2 , and the result follows. ¤ 1

pa−d+h

2.5. Higher Decomposition Numbers, Subsections In order to improve Theorem 2.4 one is led to pass to certain subgroups of X and blocks related to B. Let Y be a subgroup of X and b be a block of Y . We say that the induced block bX exists provided the map ωbX , defined P by ωbX (b c) = ωb ( x∈Y ∩c x) for any conjugacy class c of X, is an F -algebra homomorphism on the centre of F X. In this case ωbX determines a unique block B = bX of X. Example 2.5a. Suppose b is a block of the subgroup Y of X containing an irreducible character θ such that χ = IndX irreducible. Then, Y (θ) is P by formula (1.2b) for induced characters, ωχ (b c) = ωθ ( x∈Y ∩c x) for each conjugacy class c of X. Thus in this case B = bX is defined, and χ ∈ Irr(B). If Y contains DCX (D) for some p-subgroup D of X, then for any block b of Y the induced block B = bX is defined, and the defect groups of b are contained in certain defect groups of B [F, III.9.4 and 9.6]. If NX (D) ⊆ Y , then by Brauer’s First Main Theorem on blocks b 7→ bX is a bijection from the blocks of Y with defect group D to the blocks of X with defect group D [I, 15.45], or [F, III.9.7]. Let B be a block of X with defect group D, and let y ∈ Z(D) be a central element of D, say of order pn . Let Y = CX (y). Then bX is defined


27

for each p-block b of Y , and there exists b such that bX = B and D is a defect group of b [F, V.9.2]. Fix such a block b. Since y is in the centre of Y , for each irreducible character χ of X and any ϕ ∈ IBr(Y ), there exist unique dyχϕ ∈ Z[ε] ⊆ R such that X χ(xy) = dyχϕ ϕ(x) ϕ∈IBr(Y )

P for all p -elements x ∈ Y . The numbers dyχϕ = θ∈Irr(Y ) hχ, θiY · λθ (y) · dθϕ , where ResYhyi (θ) = θ(1)λθ , are called the higher decomposition numbers with respect to y. They are algebraic integers in the field of pn th roots of unity over the rationals. It follows from Brauer’s Second Main Theorem on blocks [F, IV.6.1] that if dyχϕ 6= 0 for some ϕ ∈ IBr(b), then χ belongs to bX = B. Using the second orthogonality relations (1.1b) one checks that P y y χ∈Irr(B) dχϕ dχψ is the (ϕ, ψ)-entry of the Cartan matrix Cb of b. 0

Definition 2.5b. The set of columns (dyχϕ )χ∈Irr(B) , with ϕ varying over IBr(b), is called the (major) subsection (y, b) associated to bX = B (where the term major indicates that b and B have a defect group in common). For χ, ζ in Irr(B) we define X (y,b) mχζ = dyχϕ · hϕ, ψiYp0 · dyζψ . ϕ,ψ∈IBr(b) (y,b) mχζ

(y,b) mζχ

(y,b)

By definition = and, by Theorems 2.2a and 2.3a, pd mχζ is an algebraic integer in R and is nonzero if χ is of height zero in B. Let Qb denote the quadratic form obtained from the Hermitian form defined by the positive definite symmetric matrix pd Cb−1 with integer entries. Then (y,b) by definition Qb (z) = pd mχχ if z = (dyχϕ )ϕ∈IBr(b) . Lemma 2.5c. Let (y, b) be a major subsection to B = bX , with defect d. P (y,b) Then χ∈Irr(B) mχχ = `(b) and, for each χ ∈ Irr(B), the trace TrK|Q (pd m(y,b) χχ ) ≥ [K : Q] · min Qb (z), where z = (zϕ ) varies over all nonzero vectors with integral coordinates. P In the case that ϕ∈IBr(b) dyχϕ ϕ(1) 6≡ 0 (mod p) for all χ ∈ Irr(B), it sufP fices to take the minimum over those vectors for which ϕ∈IBr(b) zϕ ϕ(1) 6≡ 0 (mod p). (y,b)

Proof. Consider the k(B) × k(B)-matrix M = (mχζ )χ,ζ . By direct computation, using that Cb−1 = (hϕ, ψiYp 0 )ϕ,ψ by Theorem 2.2a, one obtains that M 2 = M [F, V.9.4]. Since the rank of M equals the rank of pd Cb−1 ,

28


which is `(b), we infer that the trace of M is equal to `(b). It follows that P (y,b) = `(b). For the statement concerning the field traces we may χ mχχ replace K by K0 = Q(ε0 ) where ε0 is a primitive pn th root of unity such (y,b) that all mχζ ∈ K0 . Then use that TrK0 |Q (εj0 ) = −pn−1 if j is divisible by n−1 p but not by pn , and zero otherwise [F, V.9.14]. ¤ Theorem 2.5d (Brauer). Let (y, b) be a major subsection to B (defect d). (i) Assume that Qb (z) ≥ `(b) for each nonzero vector z = (zϕ ) with P integer coordinates; if ϕ∈IBr(b) dyχϕ ϕ(1) 6≡ 0 (mod p) for all χ ∈ Irr(B), P consider only those vectors for which ϕ∈IBr(b) zϕ ϕ(1) 6≡ 0 (mod p). Then k(B) ≤ pd . (ii) If B contains no irreducible character of positive height, then we p have k(B) ≤ pd `(b). (y,b)

Proof. (i) By hypothesis and Lemma 2.5c, TrK|Q (pd mχχ ) ≥ [K : Q]`(b) P (y,b) for each χ ∈ Irr(B). Using that χ∈Irr(B) mχχ = `(b) we get the estimate d p [K : Q]`(b) ≥ k(B)[K : Q]`(b). Hence the result. (y,b)

(ii) Now mχζ 6= 0 for all χ, ζ in Irr(B). Therefore, varying σ over the Galois group Γ = Gal(K|Q) and ζ over Irr(B), by the arithmetic–geometric mean inequality (1.5b) we have X Y ¡ (y,b) ¢σ 2 1 1 (y,b) 1 ≤ ( (pd |mχζ |σ )2 ) k(B)|Γ| ≤ | . p2d | mχζ k(B)|Γ| ζ,σ

ζ,σ

The term on the right equals P (y,b) that χ mχχ = `(b).

1 k(B)|Γ|

P

2d (y,b) σ σ (p mχχ ) .

Use finally once more ¤

Remark . One knows that the k(B) conjecture is true for blocks with cyclic defect groups, even for abelian defect groups of rank at most 2. The best general result so far for p-blocks B of defect d ≥ 2 is the Brauer–Feit bound k(B) ≤ p2d−2 , which follows by combining Theorems 2.4 and 2.5b [F, VII.10.13 and 10.14]. 2.6. Blocks of p-Solvable Groups X is called p-solvable provided each composition factor of X either is a p-group or a p0 -group. Then X contains a p-complement, G, and each p0 subgroup of X is contained in a conjugate of G (P. Hall; the proof is easily worked out by induction applying the Schur–Zassenhaus theorem). The block theory for p-solvable groups has been developed in [Fong, 1962].


29

A p-solvable group X is p-constrained, that is, if P is a Sylow psubgroup of Op0 p (X), then CX (P ) ⊆ Op0 p (X) (Hall–Higman lemma). Here Op (X) and Op0 (X) are the largest normal p-subgroup and p0 -subgroup of X, respectively, and Op0 p (X)/Op0 (X) = Op (X/Op0 (X)). Theorem 2.6a. Suppose X is p-solvable with Op0 (X) = 1. Then X has a unique p-block B. Assume the k(GV ) theorem has already been proved. Then k(B) ≤ |D|. We even have k(B) < |D| unless X = Z(X) × GV where G is a p-complement in X acting faithfully on the elementary abelian p-group V , and if k(GV ) = |V |. Proof. Let P = Op (X). Since Op0 (X) = 1, we know that CX (P ) ⊆ P . Hence by Theorem 2.3c the principal block is the unique p-block of X. We first show that k(X) ≤ |D| and that this inequality is proper if D is nonabelian. We argue by induction on |X|, following [Robinson, 2004]. Let G be a p-complement in X. Suppose GP 6= X. Then by induction k(GP ) ≤ |P | and either P is abelian or k(GP ) < |P |. From part (i) of Theorem 1.7a it follows that k(X) ≤ k(GP ) · |X : GP | = k(GP ) · |D : P | ≤ |D|, and that equality only holds if k(GP ) = |P |, N = GP is normal in X, k(X/N ) = |X : N | and kX (N ) = k(N ). Hence if equality holds, then P and X/N ∼ = D/P are abelian, and the p-group D/P fixes each conjugacy class of the p0 -group N/P ∼ = G. Hence D/P centralizes N/P . But D/P acts faithfully on N/P as Op (X/P ) = 1. Thus from k(X) = |D| it follows that X = GP . We therefore may assume that D = P and X = GP . Choose a minimal normal subgroup M of X, which is an elementary abelian p-group. By induction k(X/M ) ≤ |D/M | and k(X/M ) < |D/M | if D/M is nonabelian. Now the estimate (1.7b) yields that k(X) ≤ |M | · k(X/M ) ≤ |D|, and equality only holds if D/M is abelian, M is central in X and no nonidentity element of M is a commutator in X (see the beginning of Sec. 1.7). But since D is nonabelian, M = D0 has order p and is generated by commutators in D. So let D = P be abelian (and normal in X = GD). We have Z(X) = CD (G) as Op0 (X) = 1, and D = Z(X)×V where V = [D, G] by Eq. (1.6a). Let V¯ = V /U be the Frattini quotient of V , which is a faithful Fp G-module. By the k(GV ) theorem and (1.7b) k(X) ≤ |Z(X)| · |U | · k(GV¯ ) ≤ |D|,

30


and equality only holds if k(GV¯ ) = |V¯ | and U is central in X. But then U = 1 and V = V¯ , completing the proof. ¤ Theorem 2.6b (Fong). Let X be p-solvable, N = Op0 (G) and G = X/N . Let B be a p-block of X. Suppose B covers some θ ∈ Irr(N ). (i) There is a unique block b of T = IX (θ) covering θ such that bX = B, and b, B have a defect group in common. Induction defines a bijection from Irr(b) and IBr(b) to Irr(B) and IBr(B), respectively. (ii) Suppose θ is X-invariant (T = X). Then the Clifford correspondence of Theorem 1.9c, restricted to Irr(B) resp. IBr(B), describes a bijection onto Irr(B0 ) resp. IBr(B0 ) for some unique p-block B0 of the representation group G(θ) preserving decomposition numbers. In fact, Irr(B0 ) = e The defect groups of B and B0 are Sylow p-subgroups. Irr(G(θ)|θ). Proof. (i) Let χ ∈ Irr(B) and let ψ ∈ Irr(T ) occur in ResX T (χ). Then ψ ∈ Irr(T |θ) and so IndX (ψ) is irreducible by Theorem 1.8b. Thus IndX T T (ψ) = χ by Frobenius reciprocity. Note that χ(1) = |X : T |ψ(1). Let b be the block of T containing ψ. By Example 2.5a, B = bX . Let c be a defect class for B with defect group D. Then ωB (b c) 6= 0 by definition and so ωb (b c0 ) 6= 0 for some conjugacyc class c0 ⊆ c of T . By Lemma 2.3b, a defect group D0 of b is contained in a defect group of c0 , so that we may pick D0 ⊆ D. But by character degrees both b and B have the same defect. (ii) Let θb be a character of X(θ) extending θ in the sense of Definition 1.9b. Like θ this θb is irreducible as a Brauer character, and by Theorem e (where θe is a 1.9c it defines a bijection from Irr(X|θ) onto Irr(G(θ)|θ) linear character of the cyclic central subgroup Z of G(θ) of order exp (N )). The Clifford correspondence χ = θb · ζ ↔ ζ extends to Brauer characters e is a union of preserving decomposition numbers. It follows that Irr(G(θ)|θ) irreducible characters belonging to certain blocks of G(θ). We may assume that p | |G|. Then P = Op (G(θ)) 6= 1 since Z = Op0 (G(θ)) is in the centre of G(θ) (and G(θ) is p-solvable). We have e Hence CG(θ) (P ) = P × Z. There is a unique block b0 of P Z covering θ. G(θ) e Thus is the unique block covering b0 (and θ). by Theorem 2.3c, B0 = b0 e Let D be a Sylow p-subgroup of G(θ). There is a Irr(B0 ) = Irr(G(θ)|θ). e Then p - ψ(1), and there is an irreducible character ψ of D ×Z extending θ. G(θ) constituent ζ of IndDZ (ψ) whose degree is not divisible by p. Now ζ lies over θe and so ζ ∈ Irr(B0 ). Hence D is a defect group of B0 . Of course X, X(θ), G, G(θ) have isomorphic Sylow p-subgroups.

¤


31

Theorem 2.6c (Nagao). Assume that the k(GV ) theorem has been already proved. If X is a p-solvable group and B is a p-block of X with defect group D, then k(B) ≤ |D| where the inequality is proper if D is nonabelian. Proof. We proceed by induction on |X|p = pa [Nagao, 1962]. By Theorems 2.6a and 2.6b we may assume that Z = Op0 (X) is nontrivial and central in X, and that D is a Sylow p-subgroup of X (|D| = pa ). Also Irr(B) = Irr(X|θ) for some irreducible (linear) character θ of Z. Thus k(B) = |Irr(X|θ)| ≤ k(X/Z) by Theorem 1.9c. Now Op0 (X/Z) = 1, and so Theorem 2.6a applies. Hence k(X/Z) ≤ |D|, even k(X/Z) < |D| unless DZ/Z ∼ ¤ = D is abelian. 2.7. Coprime Fp X-Modules Let V be an irreducible Fp X-module. By Wedderburn’s theorem F0 = EndX (V ) is a (commutative) finite field, which we may embed into F . Viewing V = V0 as an (absolutely) irreducible F0 X-module, F0 ⊗Fp V = L σ σ V0 is the direct sum of its Galois conjugates over the prime field. Let χ ∈ IBr(X) be the Brauer character of X afforded by V0 . Let KI be the subfield of K generated by the exp (X)p0 -roots of unity, the inertia subfield of K for p (or p), and let KD be the decomposition subfield. Thus KD ⊆ KI and Gal(KI |KD ) ∼ = Gal(F |Fp ) is cyclic generated by the Frobenius automorphism. So F0 corresponds to the intermediate field K0 = KD (χ), and P we associate to V the trace character χ ˘ = TrK0 |KD (χ) = σ∈Gal(K0 |KD ) χσ . Now assume X is a p0 -group (K = KI ). Then χ is an ordinary irreducible character of X. For each τ ∈ Gal(K|Q) there is an irreducible Fp X-module V τ affording χ ˘τ = TrK0 |KD (χτ ), which is not isomorphic to V unless τ ∈ Gal(K|KD ). But V , V0 and V τ are isomorphic as G-sets. Use the fact that the F0 -dimension of F0 ⊗ CV (Y ) = CF0 ⊗V (Y ) equals hχ, ˘ 1Y iY = [F0 : Fp ]hχ, 1Y iY = hχ ˘ τ , 1Y i Y for each subgroup Y of X. Thus |CV (Y )| = |CV0 (Y )| = |CV τ (Y )|, which gives the result (cf. Secs. 1.3 and 1.4). Remark . Let X be p-solvable. The Fong–Swan theorem tells us that then every (absolutely) irreducible Brauer character of X can be lifted to an ordinary character. This is proved via Clifford theory [F, X.2.2], which even yields a p-rational lift (having its values in KI ). Arguing as above one sees that every irreducible Fp X-module has a lift affording a p-rational character.

Chapter 3

The k(GV) Problem Let p be a rational prime. We consider the situation where X = GV is the semidirect product of a finite p 0 -group G acting faithfully on an elementary abelian p-group V . Then Op 0 (X) = 1 and so V is the unique defect group of the unique p-block of X (Theorem 2.3c). The k(GV ) problem is the special case of the k(B) conjecture asking whether k(GV ) ≤ |V |, or not. 3.1.

Preliminaries

We often write the Fp G-module V additively. In this coprime situation V is completely reducible (Maschke). All irreducible characters of X = GV have degree prime to p by Theorem 1.3b and so are of height zero in the unique p-block, B. For each v ∈ V the centralizer CX (v) = CG (v)V has a unique p-block bv as well. We have bX v = B, and the subsection (v, bv ) is major. It follows from Theorem 1.5d that k(X) ≤ |V | provided there is a vector v ∈ V such that CG (v) = 1, even k(X) < |V | unless G is abelian. This may also be deduced from Theorem 2.5d. It will turn out that such regular vectors v (belonging to regular G-orbits) exist fairly often. In Theorem 3.4d below we shall see that it even suffices to find a vector with abelian point stabilizer, a result due to [Knörr, 1984]. Proposition 3.1a. Let V = V1 ⊕ V2 be the direct sum of Fp G-modules Vi , and let Gi = CG (V /Vi ). Suppose k(Gi Vi ) ≤ |Vi | and k((G/Gi )Vj ) ≤ |Vj | for i 6= j. Then k(GV ) ≤ |V |, and if k(GV ) = |V | then G = G1 × G2 and k(Gi Vi ) = |Vi | for i = 1, 2. Proof. Cleary Gi is faithful on Vi (i = 1, 2), and G/Gi is faithful on Vj for j 6= i. Consider the normal subgroups Ni = Gi Vi of X = GV , and observe that N = N1 × N2 is a normal subgroup of X too. Applying (1.7b) to the Ni yields that k(X) ≤ |V1 ||V2 | = |V |. If we have equality, then necessarily k(Gi Vi ) = |Vi | for i = 1, 2, and k(X) = k(N ) = |V |. Moreover, then every conjugacy class of X/N is good for N (Theorem 1.7a). By Brauer’s 32

The k(GV) Problem

33

permutation lemma (Theorem 1.4b) then every irreducible character θ of N is invariant in X, and each conjugacy class of X/N is good for θ. Hence X |V | = k(X) = k(X/N ) = |V | · k(X/N ) θ∈Irr(N )

by the Clifford–Gallagher formula (1.10b). Thus N = X.

¤

Each irreducible (linear) character λ of V can be extended to its ine having IG (λ) in ertia group IX (λ); indeed there is a unique extension λ its kernel. For each irreducible character θ of IG (λ), inflated to IX (λ), e the induced character χλ,θ = IndX IX (λ) (λθ) is irreducible (Theorem 1.8b). By Proposition 1.6b, V and Irr(V ) are isomorphic G-sets. We fix a Gisomorphism v 7→ λv between these G-sets, with λ0 = 1V , and we write χv,θ in place of χλv ,θ . The Clifford–Gallagher formula gives the following. P Proposition 3.1b. k(GV ) = i k(CG (vi )) where {vi } is a set of representatives of the G-orbits on V . More precisely, for each v ∈ V and θ ∈ Irr(CG (v)) the character χv,θ of X = GV is irreducible of degree |G : CG (v)| · θ(1), and these are just all the k(CG (v)) distinct irreducible characters of X lying above λv . In this manner the partition of V into G-orbits gives rise to a corresponding partition of Irr(GV ). Since the p-section of any v ∈ V in X = GV , that is, the elements in X whose p-part is conjugate to v, is the union of conjugacy classes of X represented by certain p-regular elements in CX (v) = CG (v)V , the formula on k(GV ) likewise follows from this observation (by conjugacy of the complements to V in CX (v)). (In the non-coprime situation, where G is a complement of the abelian group V in X = GV , or where just G = X/V , the corresponding formula holds replacing V by Irr(V ) and centralizers by inertia groups.) We may compute k(GV ) also as follows. Lemma 3.1c. Let {gj } be a set of representatives for the conjugacy classes P of G (1 ≤ j ≤ k(G)). Then k(GV ) = j |C`(CG (gj )|CV (gj ))|. Proof. For g ∈ G let Ωg denote the set of V -conjugacy classes contained in the coset V g. By Theorem 1.7d, |Ωg | = |CV (g)|. Since g is a p0 -element, V = CV (g) ⊕ [V, g]

34


by Proposition 1.6a, and this is a decomposition of CG (g)-modules. Of course CG (g) acts on Ωg as well (by conjugation). Now for each v ∈ V we have (vg)V = vg V = v[V, g]g. The assignment (vg)V 7→ v[V, g] is a bijection from Ωg onto V /[V, g] ∼ = CV (g) which is compatible with the action of CG (g). Hence the result. ¤ Replacing CV (g) by V /[V, g] Lemma 3.1c also holds in the non-coprime situation. 3.2. Transitive Linear Groups Suppose G is transitive on V ] = V r {0}. Fix any v ∈ V ] . Then k(GV ) = k(G) + k(CG (v)) by Proposition 3.1b. Such groups G exist. Identifying V with the additive group of a finite field and G with its multiplicative group we get a cyclic subgroup of GL(V ) of order |V | − 1 acting regularly on V ] . These Singer cycles in GL(V ) are conjugate since their generators have the same irreducible minimum polynomial over the prime field. Theorem 3.2. Suppose G is transitive on V ] . Then k(GV ) ≤ |V |, and if k(GV ) = |V | then either G is a Singer cycle in GL(V ), or pm = 23 and G is a Frobenius group of order 21, or pm = 32 and G is semidihedral of order 16. Proof. In the exceptional cases indeed k(GV ) = |V |. We shall appeal to [Hering, 1985] for the classification of the transitive linear groups (see also [Huppert–Blackburn, 1982, XII.7.5] for a summary of this work). By Theorem 1.5b we may assume that H = CG (v) is nontrivial, so that G is nonabelian (and not a Singer cycle). Let |V | = pm . Note that the semidirect product GV = V : G is a 2-transitive permutation group (of degree pm ). Embed G into GL(V ) and let F be a maximal G-invariant subfield of End(V ) containing the identity (with G acting via conjugation). Let |F | = pr so that V is a vector space of dimension m r over F . Using that |G| is not divisible by p we obtain that one of the following holds: (i) m = 2, r = 1 and p is equal to 11, 19, 29 or 59, and G contains a normal subgroup N ∼ = SL2 (5). (ii) m = 2 or 4, r = 1 (in both cases), and G contains a normal extraspecial 2-subgroup E of order 2m+1 such that CG (E) = Z(E) and G/E acts faithfully on E/Z(E). Moreover, if m = 2 then p = 3, 5, 7, 11 or 23, and if m = 4 then p = 3. (iii) m = r and G is a subgroup of ΓL1 (pm ).

The k(GV) Problem

35

∼ SL2 (5) is a maximal subgroup of the perfect group SL2 (p) as In (i) N = follows from Dickson’s list of subgroups of PSL2 (p) [Huppert, 1967, II.8.27]. Thus G/N maps injectively into GL2 (p)/SL2 (p) ∼ = F?p and has prime order. Applying the estimate (1.7b) we get k(G) ≤ k(N ) · k(G/N ) ≤ k(SL2 (5)) · (p − 1) = 9(p − 1). Since H = CG (v) is faithful and completely reducible on V , it must be cyclic of order dividing p − 1. Hence k(GV ) = k(G) + k(H) ≤ 10 · (p − 1) < p2 = |V |. Consider (ii). (Extraspecial groups are briefly discussed in Sec. 4.2 below.) Regard G as a subgroup of G0 = NGL(V ) (E). If m = 2 and p = 3 then either E ∼ = Q8 is a quaternion group, which is regular on V ] , or E ∼ = D8 ] ∼ is dihedral, which is not transitive on V . In both cases G0 = ΓL1 (32 ) is a (semidihedral) Sylow 2-subgroup of GL2 (3), and we have G = G0 . (This possibility will come up also in (iii).) For m = 2 and p ≥ 5 we must have E∼ = Q8 , for otherwise G0 would not be transitive on V ] . The cases p = 5 and p = 7 are of special interest (cf. Sec. 6.1 below): If p = 5, then G0 is a 5-complement in GL2 (5) (of order 96) and CG0 (v) is cyclic of order 4. Either G = G0 or |G0 : G| = 2, and G0 /Z(G0 ) ∼ = S4 . If p = 7, then G0 ∼ = X ◦ Z6 = X × Z3 where X is a Schur cover of S4 , and |CG0 (v)| = 3. In both cases k(GV ) ≤ k(G0 V ) < |V | by (1.10b). For m = 2 and p = 11 we have G0 ∼ = GL2 (3) × Z5 and |CG (v)| = 2 0

for any nonzero v ∈ V . Thus G = G0 by assumption, and k(GV ) < |V |. The case m = 2, p = 23 is ruled out since then G = G0 is regular on V ] . 0 Finally let m = 4, p = 3. Since |V ] | = 80 and O+ 4 (2) is a 5 -group, 1+4 5 ∼ E = 2− is extraspecial of negative type (and order 2 ) and G/E is a ∼ (30 -) subgroup of O− 4 (2) = S5 of order 5, 10 or 20 (determined up to conjugacy). Hence GV is one of the three exceptional 2-transitive solvable (Bucht) groups [Huppert–Blackburn, 1982, XII.7.4]. Here H = CG (v) is cyclic of order 2, 4 or 8, respectively. In all cases k(GV ) < |V |. The largest Bucht group again is of special interest for us (Sec. 6.1).

It remains to consider (iii). So let G be a (nonabelian) subgroup of ΓL1 (pm ), whence m ≥ 2. Then G is a subgroup of T = NGL(V ) (S for some Singer cycle S in GL(V ). This T ∼ = ΓL1 (pm ) acts on S like the Galois group of Fpm |Fp , so that T /S is cyclic of order m. In this case F = S ∪ {0} is a maximal G-invariant subfield of End(V ). Let N = G ∩ S. Then G/N is cyclic of order n, say, where n > 1 is a divisor of m. Note that n is not divisible by p. Since S is regular on V ] and G is transitive, H = CG (v) is cyclic of order dividing n. Hence k(GV ) ≤ k(G) + n.

36


Assume CT (N ) > S. Then there is a proper divisor d of m such that |N | divides pd − 1. Since G is transitive on V ] , we get that pm − 1 ≤ |G| ≤ m(pd − 1) and so m ≥ 1 + pm−d . This is impossible as d ≤ m 2 and m ≥ 2. We conclude that N = CG (N ) is irreducible on V . Just the elements in N of order dividing pm/n − 1 are central in G, and if d | n then those of order dividing pdm/n − 1 are centralized by the subgroup of G/N with index d. Hence using (1.7b) we get the (crude) estimate k(G) ≤

X n m 1 (pdm/n − 1) ≤ (pm − 1) + n(pb 2 c − 1). 2 d n d|n

Here we used that

m n

≤ bm 2 c and

P

n dm/n n6=d|n d2 (p

− 1) ≤ n

Pb m2 c j=1

pj .

Suppose we have pm = |V | ≤ k(G) + n. Then we must have n = m. In particular, m is not divisible by p. The resulting inequality pm ≤

m 1 m (p − 1) + m(pb 2 c − 1) + m m

forces that p ≤ 3, and that m ≤ 2 for p = 3, while m ≤ 4 for p = 2. We conclude that pm = 32 or pm = 23 . If G ∼ = ΓL1 (32 ) then k(G) = 7 and k(G) + m = |V |, and if G ∼ = ΓL1 (23 ) then k(G) = 5 and k(G) + m = |V | likewise. So in these two cases k(GV ) = |V |, otherwise k(GV ) < |V |. We are done. ¤ 3.3. Subsections and Point Stabilizers In this section let K, R and p|p be as in the previous chapter, but we let Γ denote the subgroup of Gal(K|Q) fixing the p0 -roots of unity in K. Since X = GV has the exponent exp (G) · p and G is a p 0 -group, Γ ∼ = Gal(Q(εp )|Q) where εp = e2πi/p . The assignment g G 7→ g X is a bijection from C`(G) onto C`(Xp0 ) (Schur–Zassenhaus). Let B be the unique pblock of X. Identifying Irr(G) with IBr(X) = IBr(B) via inflation, for ϕ ∈ Irr(G) and χ ∈ Irr(X) the decomposition number dχϕ = hχ, ϕiG . Thus P IndX G (ϕ) = χ∈Irr(X) dχϕ χ by Frobenius reciprocity (1.2c), and this may be identified with the projective character ϕ b in view of Theorem 2.2a. It follows from (2.1b) and Theorem 2.2a that the projective characters form a basis of the class functions on X vanishing off p-regular elements of X. Hence we have:

The k(GV) Problem

37

b between Lemma 3.3a. Induction of class functions yields a bijection ϕ ↔ ϕ generalized characters ϕ of G and those of X = GV vanishing off p-regular elements. For ϕ ∈ Z[Irr(G)] and g ∈ G we have ϕ(g) b = |CV (g)|ϕ(g). Proof. It remains to verify the last statement. Let ϕ ∈ Z[Irr(G)]. By formula (1.2b) for induced characters (and class functions), noting that V is a (right) transversal to G in X, we have X

ϕ(x) b = v∈V

ϕ(vxv −1 ) = |CV (g)|ϕ(g)

:vxv −1 ∈G

if x ∈ X is conjugate to some g ∈ G, and ϕ(x) b = 0 otherwise.

¤

Let πV denote the permutation character of G on the set V (as usual). Let ϕ, ψ be in Irr(G). Then Lemma 3.3a and Frobenius reciprocity tell us that the corresponding Cartan invariant of B is given by b X = hϕπV , ψiG . cϕψ = hϕ, b ψi Let CB = (cϕψ ) be the Cartan matrix. By Lemma 2.1a the class function on X = GV taking the value |V | on p-regular elements, and the value zero otherwise, is a generalized character of X. Thus by Lemma 3.3a (3.3b)

δV = |V |/πV

is a generalized character of G (δV (g) = |V : CV (g)| for g ∈ G). This generalized character has been introduced (and studied) by [Knörr, 1984]. −1 It follows from Theorem 2.2a that the (ϕ, ψ) entry of |V |CB , viewing ϕ, ψ as characters of X/V , is given by |V | X 1 X ϕ(V x)ψ(V x) = ϕ(g)|V : CV (g)|ψ(g) = hϕδV , ψiG , |X| |X| x∈Xp0

g∈G

because each element of Xp0 is conjugate to an element of G and δV (g) is the number of conjugates of g ∈ G lying in the coset V g. We apply this to HV where H = CG (v) for some v ∈ V .

38


Theorem 3.3c (Knörr). Suppose there is v ∈ V such that for H = CG (v) we have hθδV , θiH ≥ k(H) for all θ ∈ Z[Irr(H)] with θ(1) 6≡ 0 (mod p). Then k(GV ) ≤ |V |, and equality only holds if hχδV , χiH = k(H) for all χ ∈ Irr(G). Proof. Let Cv be the Cartan matrix of the unique p-block bv of Y = HV , and let Qv be the quadratic form associated to |V |Cv−1 . As seen above the (ϕ, ψ) entry of |V |Cv−1 is given by hϕδV , ψiH . Of course k(H) = `(bv ). Let (v, bv ) be the corresponding (major) subsection to the block B. (v,b ) Define mχζ v as in Sec. 2.5. For each irreducible character χ of X = GV we have χ(v) ≡ χ(1) 6≡ 0 (mod p) by Theorem 1.3b. Hence X

dvχϕ ϕ(1) = χ(v) 6≡ 0 (mod p).

ϕ∈Irr(H)

Thus k(X) ≤ |V | by hypothesis and Theorem 2.5d, and this equality is (v,b ) proper provided TrK|Q (|V |mχχ v ) > [K : Q] · k(H) for some χ ∈ Irr(X). (v,b )

Now let us investigate mχχ v when χ ∈ Irr(G) (inflated to X). Then the higher decomposition number dvχϕ = hχ, ϕiH is nothing but the multiplicity of ϕ ∈ Irr(H) in the restriction to H of χ. Thus v) |V |m(v,b = χχ

X

hχ, ϕiH hϕδV , ψiH hχ, ψiH = hχδV , χiH .

ϕ,ψ∈Irr(H)

This completes the proof.

¤

The reader is referred to [Knörr, 1984] for a proof of the above result avoiding block theory. Knörr noticed that the hypothesis in Theorem 3.3c is fulfilled if there is a generalized character ψ of H = CG (v) such that ψ(h) ∈ p for h ∈ H ] but ψ(1) 6∈ p, and such that δV ≥ |ψ|2 on H (elementwise). Then hθδV , θiH ≥ hθψ, θψiH for each θ ∈ Z[Irr(H)] with p - θ(1), and |H|hθψ, ζiH =

X

¯ θ(h)ψ(h)ζ(h) ≡ θ(1)ψ(1)ζ(1) (mod p)

h∈H

for each ζ ∈ Irr(H), where p - ζ(1) by Theorem 1.3b. So each ζ ∈ Irr(H) is contained in θψ, which gives the result. Following [Robinson–Thompson, 1996] we use the following slightly different concept, which will turn out to be fulfilled when v is a so-called real vector for G.

The k(GV) Problem

39

Theorem 3.3d (Robinson–Thompson). Let H = CG (v) for some v ∈ V . Assume there is a faithful Fp H-submodule U of V and a rational-valued generalized character ψ of H such that ψ 2 = δU on H, or p is odd and ψ 2 = δU on a subgroup N of H with |H : N | = 2, while ψ(h)2 = p1 δU (h) for h ∈ H r N . Then k(GV ) ≤ |V |, even k(GV ) ≤ p+3 2p |V | in the latter (odd) case, and in both cases the inequalities are proper unless [V, H] ⊆ U . Proof. Let Y = HV , and observe that hvi is in the centre of Y . Suppose σ ∈ Γ sends εp to εsp . Then for each h ∈ H and each irreducible character χ of X (or Y ) we have χσ (hv) = χ(hv)τ = χ(hv s ) (choosing s ≡ 1 (mod |G|) and representing hv = vh by a diagonal matrix with trace χ(hv)). Suppose first that ψ 2 = δU on H. Then ψ(h) is divisible by p for all h ∈ H ] , as H is faithful on U , and ψ(1) = ±1. As in Lemma 3.3a let b ζi for ζ ∈ Irr(Y ). Define ψb = IndYH (ψ), and let nζ = hψ, ψe =

X ζ∈Irr(Y )

nζ

ζ(v −1 ) ζ. ζ(1)

e So ψe and Ψ = IndX Y (ψ) are pth cyclotomic integer combination of characters of Y and X, respectively. Hence hΨ, χi ∈ Z[εp ] for each χ ∈ Irr(X). We e have ψ(y) = |CV (h)|ψ(h) if y ∈ Y is conjugate to hv for some (unique) e h ∈ H, and ψ(y) = 0 otherwise. Since there are |V : CV (h)| · |H : CH (h)| elements in Y conjugate to hv for each h ∈ H, by Frobenius reciprocity X e χiY = ψ(h)χ(h−1 v −1 ) ≡ χ(v −1 ) (mod p) |H|hΨ, χiX = |H|hψ, h∈H

for each χ ∈ Irr(X). Thus hΨ, χi 6= 0 as p - χ(1) by Theorem 1.3b. Moreover, hΨ, χσ i = hΨ, χiσ 6= 0 for each σ ∈ Γ. Hence by the arithmetic– P geometric mean inequality (1.5b) σ |hΨ, χσ i|2 ≥ p − 1. Similarly Ψ(x) = |CV (h)|ψ(h) if x ∈ X is conjugate to hv for some h ∈ H, and zero otherwise. It follows that X e ψi eY = 1 |CV (h)|ψ(h)2 . hΨ, ΨiX = hψ, |H| h∈H

Now ψ 2 = δU on H, and δV = δU · δU 0 if V = U ⊕ U 0 as an H-module. We conclude that hΨ, Ψi ≤ |V | and that this inequality is proper unless δU 0 = 1H , that is, [V, H] ⊆ U . This gives the result in the first case, as P P (p − 1)k(X) ≤ σ∈Γ χ∈Irr(X) |hΨ, χσ i|2 = (p − 1)hΨ, Ψi ≤ (p − 1)|V |.

40


Let now p be odd and ψ 2 = δU on N but ψ(h)2 = p1 δU (h) for h ∈ H rN (|H : N | = 2). Let µ be the linear character of H with kernel N , and let ϕ = ψ + µψ. Then ϕ(h) = 2ψ(h) for h ∈ N , while ϕ vanishes on H r N . Hence ϕ(h) is an integer multiple of p for each h ∈ H ] , while ϕ(1) = 2ψ(1) is not divisible by p. Like above we define the class function Φ on X by letting Φ(x) = |CV (h)|ϕ(h) if x is conjugate to hv for some h ∈ H. Then, as before, hΦ, χσ i = hΦ, χiσ 6= 0 for each χ ∈ Irr(X) and P σ 2 σ ∈ Γ, so that σ |hΦ, χ i| ≥ p − 1 by (1.5b). Write Φ = Φ0 + Φ1 , i where Φi (x) = µ (h)|CV (h)ψ(h) if x is conjugate to hv for some h ∈ H, and 0 otherwise. Then at least one of hΦi , χi 6= 0, i = 0, 1, and then hΦi , χσ i = hΦi , χiσ 6= 0 for each σ ∈ Γ. Let Si be the set of irreducible characters of X = GV which occur P with nonzero multiplicity in Φi . If χ ∈ Si then σ |hΦi , χσ i|2 ≥ p − 1 by (1.5b). It follows that (p − 1)|Si | ≤

X X

|hΦi , χσ i|2 = (p − 1)hΦi , Φi i,

σ∈Γ χ∈Si

where hΦi , Φi i =

1 |H|

P h∈H

|CV (h)|ψ(h)2 is the same for i = 0, 1 and equals

1 1 X |CV (h)|δU (h) + |H| |H| h∈N

X

|CV (h)|δU (h)/p ≤

h∈HrN

|V | |V | + . 2 2p

Thus |Si | ≤ p+1 2p |V | for i = 0, 1 and, as above, these inequalities are proper unless [V, H] ⊆ U . We also see that hΦ0 − Φ1 , Φ0 − Φ1 i =

1 |H|

X h∈HrN

4|CV (h)|δU (h)/p ≤

2 |V |. p

Hence at most p2 |V | irreducible characters of X occur with nonzero multiplicity in Φ0 − Φ1 , and again we only can have equality when [V, H] ⊆ U . We see that |S0 r S1 | + |S1 r S0 | ≤ p2 |V | (arguing as above considering the Γ-classes in Irr(X)). Now k(GV ) = |S0 ∪ S1 | = 21 (|S0 | + |S1 | + |S0 r S1 | + p+3 2 |S1 r S0 |) ≤ |V2 | (2 · p+1 ¤ 2p + p ) = 2p |V |, completing the proof.

The k(GV) Problem

41

3.4. Abelian Point Stabilizers Suppose H 6= 1 is an abelian p0 -subgroup of GL(V ) = GLm (p). If H is irreducible, then F = EndH (V ) is a finite field containing H. Hence H = hyi is cyclic, F = Fp [y] and CGL(V ) (H) is a Singer cycle in GL(V ). So |H| divides pm − 1, but |H| does not divide pn − 1 for 1 < n < m. Also, H acts semiregularly on V ] . In this irreducible case define δH = (|H| + 1)1H − ρH where ρH is the regular character of H. If H is a Singer cycle, then δH = δV is Knörr’s generalized character (on H). Otherwise let t ≥ 2 denote by the number of H-orbits on V ] . Then k(HV ) = t + |H| = t + |V |−1 t Proposition 3.1b. From 1 < t < |V | − 1 we get that k(HV ) < |V | and that δV = (t|H| + 1)1H − tρH ≥ (|H| + 1)1H − ρH = δH at each element of H, both functions taking only nonnegative real values. Of course hδV , 1H i = t(|H| − 1) + 1 > |H| = hδH , 1H i by the choice of t. Lemma 3.4a. Suppose H 6= 1 is an abelian p0 -subgroup of GL(V ), and let Ln [V, H] = i=1 Vi be a decomposition into irreducible Fp H-modules (so that n ≥ 1 and no Vi is a trivial module). Let Hi = H/CH (Vi ) for each i, and Q define δH = i δHi , which is a generalized character of H. We have δV ≥ δH at each element of H, and hδV , 1H i > hδH , 1H i unless δV = δH and each Hi is a Singer cycle in GL(Vi ). Proof. By Proposition 1.6a, V = CV (H) ⊕ [V, H], and [V, H, H] = [V, H]. Q It is obvious that δV = i δVi . The result follows. ¤ We mention that H has a regular orbit on V , because picking arbitrary P vi ∈ Vi] for each i and letting v = i vi , then (3.4b)

CH (v) =

T i

CH (vi ) =

T

Hi = 1.

42


Proposition 3.4c. Keep the assumptions of the preceding lemma, embed each Hi uniquely into the corresponding Singer cycle Gi on Vi and let G = Q i Gi be the direct product. Then H embeds into G in a natural way. We have hθδV , θiH ≥ |H| for each nonzero generalized character θ of H, and if hδV , 1H iH = |H| then necessarily H = G. Proof. Of course G is a p0 -group and H is a certain fibre-product of the Hi Q and so embeds into the direct product i Hi , which is a subgroup of G. For the first statement in the lemma it suffices to show that hθδH , θi ≥ |H| for any generalized character θ 6= 0 of H (which will be fixed in what follows). Let us write H ∗ = Hom(H, C? ) for the character group. Restriction from G to H defines an epimorphism from the character group G∗ = Qn ∗ (direct) onto H ∗ . For each subset I ⊆ N = {1, 2, · · ·, n} let i=1 G Qi ∗ GI = i∈I G∗i (direct), and let HI∗ be its image in H ∗ . Here each λ ∈ G∗I Q is sent to λI = i∈I λ(i) where λ(i) is the restriction of the ith component of λ to Hi . Define Y X γλ = (1H − λ(i)) = (−1)|J| λJ . i∈I

J⊆I

P P We assert that δ = δH = I⊆N 2−|I| λ∈G∗ |γλ |2 . By Lemma 3.4a this is I true for n = 1 (since by definition γ∅ = 1H ). Proceeding by induction on n, P the assertion follows. Using that hθ|γλ |2 , θi = hθγλ , θγλ i = χ∈H ∗ hθγλ , χi2 we obtain that X X hθδ, θi = 2−|I| hθ, γλ χi2 . I⊆N

(λ,χ)∈G∗ ×H ∗ I

We define a map H ∗ → P(N ) assigning to each χ ∈ H ∗ a subset I of N of smallest cardinality, arbitrarily chosen, such that hθ, λI χi 6= 0 for some λ ∈ G∗I . Such subsets exist since θ 6= 0 and H is faithful on [V, H]. Let M (I) = M (I, θ) denote the inverse image in H ∗ of the subset I of N with respect to this map. For χ ∈ M (I) we have hθ, γλ χi =

X

(−1)|J| hθ, λJ χi = (−1)|I| hθ, λI χi

J⊆I

for all λ ∈ G∗I , and there is λ ∈ G∗I such that hθ, λI χi 6= 0. Clearly H ∗ is the disjoint union of the fibres M (I), and it suffices to show that P 2−|I| (λ,χ)∈G∗ ×H ∗ hθ, γλ χi2 ≥ |M (I)| for all I. I

The k(GV) Problem

43

Fix I ⊆ N , and let ϕ ∈ H ∗ . The Boolean group B = P(I), which is an elementary abelian 2-group with respect to the symmetric difference J ⊕J 0 = (J ∪J 0 )r(J ∩J 0 ), acts on the set G∗I ×ϕHI∗ via (µ, ζ)J = (µJ , µJ ζ) where µJ ∈ G∗I is defined by µJ (i) = µ(i)−1 for i ∈ J and µJ (i) = ζ(i) 0 0 otherwise. Indeed (µJ )J = µJ⊕J and µJ (µJ )J 0 = µJ⊕J 0 . If (λ, χ) = J (µ, ζ) is in the B-orbit of (µ, ζ) then γλ χ = γµJ µJ ζ = µ

Y j∈J

Y

µ(j)

(1 − µ(j)−1 )

j∈J

Y

(1 − µ(i)) = (−1)|J| γµ ζ,

i∈IrJ

so that hθ, γλ χi2 = hθ, γµ ζi2 . Moreover, if C is the stabilizer in B of (µ, ζ) then |C|−1 γµ is a generalized character of H. In fact, for J ∈ C and J 0 ∈ B we have µJ⊕J 0 = µJ and |J ⊕ J 0 | ≡ |J| + |J 0 | (mod 2), whence P P γµ = J⊆I µJ = J∈C (−1)|J| τ for some generalized character τ of H. If not all J ∈ C have even cardinality, those of even cardinality form a P subgroup of C of index 2 and then J∈C (−1)|J| = 0. It follows that 2−|I|

X

hθ, γλ χi2 = |C|−1 hθ, γµ ζi2

(λ,χ)∈(µ,ζ)B

is a nonnegative rational integer. Consider the set M (I) ∩ ϕHI∗ , which can be empty. We assert that the nonnegative integer 2−|I|

X

hθ, γλ χi2 ≥ |M (I) ∩ ϕHI∗ |.

(λ,χ)∈G∗ ×ϕHI∗ I

Varying over the subsets I of N and the cosets of H ∗ mod HI∗ this will show that hθδ, θi ≥ |H|. We may assume that M (I)∩ϕHI∗ 6= ∅. Let χ = χ1 , ···, χr be its distinct elements. By construction there is λ = λ1 ∈ G∗I such that hθ, γλ χi = (−1)|I| hθ, λI χi 6= 0. Since each χχ−1 ∈ HI∗ we find αj ∈ G∗I such that j −1 (αj )I = χχj , and we put λj = λαj for j = 2, · · ·, r. We have hθ, γλj χj i = (−1)|I| hθ, (λj )I χj i = (−1)|I| hθ, λI χi 6= 0. It now suffices to prove that the (λj , χj ) belong to different orbits under B = P(I). Suppose (λj , χj ) = (λ, χ)J = (λJ , λJ χ) for some J ⊆ I. Then −1 ∗ (αj )I = χχ−1 j = λJ and so the image of λj = λαj in HI is the identity on ∗ HJ . From χj ∈ M (I) we can conclude that J = ∅. Hence λj = λJ = λ,

44


αj = 1 and χj = χ. Replacing χ by χi we have λj = λi βj , where βj = αj αi−1 satisfies (βj )I = χi χ−1 j . So the same argument applies. Consider finally the case where θ = 1H . Then the map H ∗ → P(N ) is uniquely determined, assigning to χ ∈ H ∗ the support of χ, the smallest subset I of N such that χ ∈ HI∗ . Suppose we have hδV , 1H iH = |H|. By Lemma 3.4a this forces that ResG H (δV ) = δ and that Hi = Gi for each i. Assume that H 6= G (so that there must be a certain amalgamation). Then there is a subset I ⊆ N such that the map G∗I → HI∗ is not injective. Consider the set M (I) = M (I, 1H ) of characters in H ∗ with support I. Then ∅ 6= M (I) ⊆ HI∗ . Keep the notation of the preceding paragraph, picking ϕ ∈ HI∗ . Hence for χ = χ1 in M (I) we have λI = χ−1 in HI∗ , and so on. By assumption there is µ 6= λ in G∗I with µI = λI . It remains to show that the B-orbit of (µ, χ) is different from the orbits of all (λj , χj ), because this will yield the desired contradiction hδ, 1H i > |H|. If (µ, χ)J = (λj , χj ) for some J ⊆ I, then χj = µJ χ = λJ χ and µJ = λj . Pick τ ∈ G∗IrJ such that τIrJ = λIrJ . Then h1H , τIrJ χj i = h1H , λIrJ λJ χi = h1H , λI χi 6= 0. We conclude that I r J = I, that is, J = ∅. Hence χj = λJ χ = χ and µ = µJ = λj = λ, against our choice. This completes the proof. ¤ Theorem 3.4d (Knörr). Suppose there is v ∈ V such that H = CG (v) is abelian. Then k(GV ) ≤ |V |, and we have equality only if k(HV ) = |V | in which case HV is the direct product of certain Hi Vi where Hi either is a Singer cycle on Vi or Hi = 1, |Vi | = p. Proof. This is immediate from the preceding proposition and Theorem 3.3c. ¤ Unfortunately the search for abelian vectors v ∈ V , for which CG (v) is abelian, is not compatible with Clifford reduction. So Theorem 3.4d will not be relevant for the proof of the k(GV ) theorem. We shall make use of this theorem when discussing the question under which conditions we can have equality k(GV ) = |V | (Chapter 11).

Chapter 4

Symplectic and Orthogonal Modules The ultimate target of this chapter is to show that the assumptions made in Theorem 3.3d are fulfilled if the module carries a nondegenerate G-invariant symplectic or orthogonal form. This makes necessary a discussion of selfdual modules and of automorphism groups (holomorphs) of extraspecial groups, which in turn lead us to a useful new concept of goodness for conjugacy classes, and to Weil characters. 4.1. Self-dual Modules Let V be a coprime F G-module, where F = Fr is a finite field of characteristic p. From Proposition 1.6b we know that V is isomorphic to Irr(V ) = Hom(V, C? ) as a G-set. But usually we do not have an isomorphism of G-modules. The F G-module V ∗ = HomF (V, F ), with diagonal action λx (v) = λ(vx−1 ) for λ ∈ V ∗ , x ∈ G, v ∈ V , is called the dual module to V . The module V is self-dual provided V ∼ = V ∗ (as F G-modules). Lemma 4.1a. V is self-dual if and only if its Brauer character is realvalued. Proof. Let ρ be a matrix representation of G on V (to some basis of V ). Then the matrix representation of G on V ∗ with respect to the dual basis is contragredient to ρ, sending x ∈ G to ρ(x−1 )t . So if V is self-dual, its Brauer character is real-valued (Sec. 1.5). For the converse use that G is a p0 -group and so the Brauer character is an ordinary character (determining the isomorphism type of V ). ¤ So the dual of a matrix representation ρ of G is obtained by applying ρ followed by the inverse transpose automorphism of the linear group. If G is a real group, every representation of G is self-dual. The F G-module V is self-dual if and only if V carries the structure of a nondegenerate G-invariant F -bilinear form, and in the coprime situation these forms are symplectic or symmetric or orthogonal sums of those forms [Gow, 1993]: 45

46


Theorem 4.1b (Gow). Let V be self-dual. Then V = U ⊕ W where U is a symplectic F G-module (with even F -dimension), affording a G-invariant nondegenerate symplectic form, and where W is an orthogonal F G-module, affording a G-invariant nondegenerate symmetric bilinear form. The action of G on W can be chosen to be trivial when p = 2. Proof. Without loss of generality V 6= 0. By hypothesis there exists an isomorphism ϕ : V → V ∗ of F G-modules. Define [v, w] = [v, w]ϕ = ϕ(v)(w) for v, w ∈ V . This is a nondegenerate F -bilinear form on V . For x ∈ G we have [vx, wx] = ϕ(vx)(wx) = ϕ(v)x (wx) = ϕ(v)(wxx−1 ) = [v, w]. Hence the form is G-invariant. The form (v, w) 7→ [v, w]+[w, v] is symmetric and G-invariant, its radical being an F G-submodule of V . Suppose first that V is an irreducible F G-module. We assert that V is a symplectic F G-module or that p 6= 2 and V is an orthogonal F Gmodule. The above symmetric form is nondegenerate or is zero. If it is zero and p 6= 2, the form [·, ·] is symplectic (alternating), and we are done. So let p = 2. The symmetric form is now symplectic, too, and we may thus assume that it is zero. Then [·, ·] is symmetric and the radical V0 = {v ∈ V | [v, v] = 0} is an F G-submodule 6= 0 of V of codimension at most 1. It follows that V0 = V and that our form is symplectic. In the general situation let U be a (proper) irreducible F G-submodule of V . If the form [·, ·] is nonzero on U , it is nondegenerate, as U is irreducible. Then U is self-dual, and the preceding paragraph applies. Otherwise U is contained in U ⊥ = {v ∈ V | [v, U ] = 0}, which is an F G-module, and we have dim F V = dim F U + dim F U ⊥ as the form is nondegenerate. By Maschke V = U ⊥ ⊕ U0 for some F G-module U0 . The map λ : U → U0∗ given by λ(u)(u0 ) = [u, u0 ] is an F G-module homomorphism. Its kernel cannot be U for otherwise we had U0 ⊆ U ⊥ . Hence λ is injective, and it is surjective since dim F U = dim F U0 = dim F U0∗ . Therefore U ∼ = U0∗ , whence U ∗ ∼ = (U0∗ )∗ = U0 . The form on U ⊕ U0 , given by (u + u0 , u0 + u00 ) 7→ [u0 , u0 ] − [u00 , u], is symplectic and G-invariant. One verifies that this is nondegenerate using again that U0 6⊆ U ⊥ . The result follows by induction on dim F V. ¤

Symplectic and Orthogonal Modules

47

Lemma 4.1c. Suppose V is a faithful irreducible F H-module for the abelian group H, and let n = dim F V. If n = 1 and V is self-dual (as an F H-module), then |H| = 1 or 2. Let n ≥ 2. Then V is self-dual if and only if n = 2m is even and |H| a divisor of rm + 1. Proof. From Sec. 3.4 we know that H = hyi is cyclic and that CGL(V ) (H) is a Singer cycle in GL(V ). We may identify V = Frn = Fr [y]; |H| is a divisor of rn − 1 but not of ri − 1 for 1 < i < n. Clearly y ∈ Frn is an eigenvalue of the F -linear map on the extension field given by multiplication with i y. So the eigenvalues of y on V are the different conjugates y r of Frn |Fr (0 ≤ i ≤ n − 1). Suppose V is a self-dual F H-module. Then for each eigenvalue the inverse must be an eigenvalue too. If n = 1 then necessarily y = y −1 and |H| = 1 or 2. Let n > 1. Then |H| > 2, and no eigenvalue j to y is equal to its inverse. Hence n = 2m is even, and if y −1 = y r for some j ≤ n − 1, then |H| is a divisor of rj + 1. It follows that |H| is a divisor of r2j − 1 and so 2j = n = 2m. We conclude that |H| is a divisor of rm + 1. Conversely, if n = 2m is even and |H| is a divisor of rm + 1, then m y −1 = y r . Then for each eigenvalue to y the inverse is an eigenvalue too, whence V is a self-dual F H-module. ¤ It follows that V is a self-dual module for a Singer cycle in GL(V ) = GLn (r) if and only if rn = 2 or 3. Since GL(V ) contains Singer cycles, the standard module V is self-dual for GL(V ) only when n = 1, r = 2 or 3, or when n = 2, r = 2. It is self-dual for SL(V ) only when n = 1 or 2, because the intersection of a Singer cycle in GL(V ) with SL(V ) is irreducible and has order (rn − 1)/(r − 1). One knows that the symplectic group Sp2m (r) and the orthogonal group O− 2m (r) have unique conjugacy classes of Singer cycles, which by definition are cyclic and irreducible. They have order rm + 1. 4.2. Extraspecial Groups Let q be a prime (usually q 6= p). A finite nonabelian q-group E is called extraspecial if the centre Z = Z(E) has order q and U = E/Z is elementary abelian, hence a vector space over Fq . Thus E is a central product of nonabelian groups of order q 3 . Let |E| = q 1+2m . Groups q1+2m : Let q be odd and E = Ω1 (E), i.e., exp (E) = q. 1+2m Then E ∼ is determined by its order. Choosing a gen= q 1+2m = q+ erator z of Z, E has the generators xi , x∗i for i = 1, · · ·, m, satisfying

48


the relations [xi , x∗i ] = z, z q = xqi = (x∗i )q = 1 for all i, the other generators centralizing each other (and z is central). The commutator map (xZ, yZ) 7→ [x, y] is a (well-defined) nondegenerate symplectic Fq -form on U , identifying Z with Fq . Every symplectic transformation on U can be lifted to an automorphism of E centralizing Z. Let A = CAut(E) (Z) and C(E) = A/Inn(E) = COut(E) (Z). We have Aut(E) = A : hαi where α permutes the elements in Z ] transitively, induces on U ∼ = Inn(E) a symplectic similitude of order q − 1 and on C(E) ∼ = Sp2m (q) the unique outer (diagonal) automorphism of order 2. In order to see this, let a be a generator of F?q and define α by xi 7→ xai , x∗i 7→ x∗i (i = 1, · · ·, m) and z 7→ z a . This preserves the defining relations and generates CSp2m (q)/Sp2m (q). Groups 21+2m : Let q = 2. Then either E ∼ is the central product = 21+2m ± + 1+2m of m dihedral groups of order 8, or E ∼ is the central product of m = 2− quaternion groups Q8 . (Observe that D8 ◦D8 ∼ = Q8 ◦Q8 .) The squaring map xE 0 7→ x2 is a (well-defined) nonsingular quadratic form Q on U = E/Z giving, together with the commutator map, a system of defining relations on E. Thus C(E) = Out(E) ∼ = O(U, Q), that is, C(E) ∼ = O+ 2m (2) if E is of − ∼ positive type and C(E) = O2m (2) otherwise. (This refers to the fact that U has Witt index m (plus) or Witt index m − 1 (minus); cf. Appendix ± ∼ B.) Except for O+ 4 (2) = S3 wr S2 there is a unique subgroup Ω2m (2) in ± O2m (2) with index 2, the kernel of the Dickson invariant. The extension Inn(E) ½ Aut(E) ³ C(E) is nonsplit when m > 2 (Appendix A10). Groups 21+2m : We have 21+2m ◦ Z4 ∼ ◦ Z4 , and the isomorphism = 21+2m + − 0 1+2m type of this group is written 20 . If E ∼ is such a group of = 21+2m 0 Z . Let A = CAut(E) (Z) and extraspecial type, its centre Z = Z(E) ∼ = 4 C(E) = A/Inn(E). An argument similar to before shows that C(E) ∼ = Sp2m (2) via the commutator map on U = E/Z. We have Aut(E) = A : hai where α is the noninner central automorphism of E, which inverts the elements of Z and centralizes U ∼ = Inn(E) and C(E). Both U ½ A ³ C(E) and U ½ Aut(E) ³ C(E) × hαi do not split for m ≥ 2 (Appendix A10). ∼ q 1+2m to indicate that E is one of the above groups. We write E = ±,0 These groups are, up to cyclic central factors, the q-groups of symplectic type all of whose characteristic abelian subgroups are cyclic and central (P. Hall); cf. [Aschbacher, 1986, (23.9)]. The character theory of E is easy. E has φ(|Z|) (Euler function) faithful irreducible characters which are determined on Z = Z(E) and vanish outside the centre. They have value field Q(e2πi/|Z| ), degree q m and are


49

conjugate under field and group automorphisms. Their Schur index is 1 except when E ∼ , in which case it is 2 and then each field in which = 21+2m − −1 is a sum of two squares is a splitting field [I, 6.18 and 10.16]. 4.3. Holomorphs 1+2m Let E ∼ , and let θ be a faithful irreducible character of E. A group = q±,0 X is called a weak holomorph of E provided E is a normal subgroup of X such that CX (E) = Z(E) = Z(X) and X/E ∼ = C(E) is the resulting (full) symplectic resp. orthogonal group on U = E/Z(E). Then X is determined by E up to isoclinism, and E is the unique minimal nonabelian normal subgroup of X unless q = 2, m = 1. By definition θ is stable in X, and we call X a holomorph of E if, in addition, there is χ ∈ Irr(X) extending θ. Then χ is faithful, of degree χ(1) = θ(1) = q m . Since the faithful irreducible characters of E are algebraically conjugate, this does not depend on the choice of θ. By stable Clifford theory each weak holomorph is a holomorph if the Schur multiplier of C(E) is trivial. ∼ 31+2 (where it has order Since C(E)/C(E)0 is of order 1 or 2 unless E = + 1+4 3) or E ∼ (where it is elementary of order 4), the faithful irreducible =2 +

characters of degree q m of a holomorph X of E have the same value field. We say that the holomorph X is standard if this character field is as small as possible. It will turn out that standard holomorphs exist and that their isomorphism type is determined by this field (which is the qth cyclotomic field for odd q and contained in the 8th otherwise). Our approach is based on work by [Griess, 1973], [Isaacs, 1973] and [Schmid, 2000]. 1+2m Proposition 4.3a. Let X be a standard holomorph of E ∼ . Let = q±,0 F be a field of prime characteristic p 6= q such that the character field of X fits into F . Let V be an F E-module affording θ as a Brauer character, embed E into GL(V ) through θ and let G0 = NGL(V ) (E). Then G0 = X ◦ Z is a central product over Z(E) = Z(X) where Z = CG0 (E) ∼ = F ?.

Proof. Let χ ∈ Irr(X) extend θ. This χ is irreducible and faithful as a Brauer character mod p as it is θ (p 6= q). Since Schur indices are 1 in prime characteristic, there is an F X-module affording χ as a Brauer character, and this may be identified with V . We may embed X into G0 through χ. Since E is absolutely irreducible on V , CG0 (E) = Z ∼ = F ? consists of scalar multiplications. Of course, G0 induces on E only automorphisms centralizing Z(E) = Z ∩ E. Hence the result. ¤

50


Examples 4.3b. (i) Let E = D8 ∼ = 21+2 + . Then the dihedral and semidihedral groups of order 16 are the unique (weak) holomorphs X of E, up to isomorphism. The two faithful X have Schur √ irreducible characters of √ index 1 and the value field Q( 2) if X is dihedral and Q( −2) otherwise. ∼ 21+2 . Then GL2 (3) = 2+ S4 and (the binary (ii) Let E = Q8 = − octahedral group) 2− S4 are the unique (weak) holomorphs of E. These groups are the two Schur covers of S4 . (We use the Atlas notation, writing 2+ Sn for that extension in which the transpositions of Sn lift to involutions.) The two faithful irreducible characters of degree 2 of 2+ S4 have Schur index √ √ 1 and value field Q( −2), those of 2− S4 have index 2 and value field Q( 2). (iii) Let E = Q8 ◦ Z4 ∼ . Embed E into GL2 (5) via some faithful = 21+2 0 irreducible character, and let X = NGL2 (5) (E). We have seen in Theo(2) rem 3.2 that X is transitive on the nonzero vectors of V = F5 . It is a 5-complement in GL2 (5), unique up to conjugacy (|X| = 96). The representation of the 50 -group X on V can be lifted to a Q(i)-representation. Hence X is a standard holomorph of E, and is the unique one up to isomorphism. (iv) Let E = 31+2 + . The symplectic group Sp4 (3) has a subgroup X = E : Sp2 (3) mapping onto a maximal parabolic subgroup of PSp4 (3) ∼ = Ω− (2). Here the centre Z = Z(X) is generated by elements in the classes 6 3A0 B0 , and |X/X 0 | = 3 [Atlas, p. 26]. The image of X is contained in a subgroup of type E : GL2 (3) of PSp4 (3).2 ∼ = O− 6 (2) which is nontrivial on Z. So there is an automorphism α of X inverting the elements of Z. Note that GL2 (3) and SL2 (3) have trivial Schur multiplier, and that all their cohomology groups on the standard module vanish. Thus X is a weak holomorph for E. Up to conjugacy under E, there is a unique complement S∼ = Sp2 (3) to E in X which is stable under α. Let ξ be one of the two faithful irreducible characters of Sp4 (3) of degree 4, which are fused by α [Atlas]. Evidently ResX (ξ) = χ + λ for some faithful irreducible character χ of X (of degree 3) and some linear character √ λ. We have Q(χ) = Q(ξ) = Q( −3). Hence X is a standard holomorph of E. There are two further holomorphs of E isoclinic to X. However, these further holomorphs appear in X ◦ Z9 and their character value fields require the 9th roots of unity. (Cf. [Atlas, p. xxiii], and observe that χ does not vanish on all elements of X r X 0 .) Thus X is the unique standard holomorph, up to isomorphism. Each faithful irreducible character of E has |X/X 0 | = 3 extensions to X, but these are conjugate under central automorphisms corresponding to Hom(X/X 0 , Z(X)).


51

1+2m Theorem 4.3c. Let E = q+ , q odd. Up to isomorphism, there is a unique standard holomorph X for E and its value field is K = Q(e2πi/q ). There is an automorphism α of X permuting the nontrivial elements in Z = Z(E) transitively and leaving invariant a unique, up to conjugacy under E, subgroup S ∼ = Sp2m (q) complementing E in X. This α induces on S the outer (diagonal) automorphism of order 2. The faithful irreducible characters of X of degree q m have Schur index 1 and are conjugate under automorphisms of X.

Proof. By the above we may assume that (m, q) 6= (3, 1). Then Sp2m (q) is perfect, and its multiplier is trivial by (A6). Let A = CAut(E) (Z). Recall from Sec. 4.2 that A/Inn(E) = C(E) ∼ = Sp2m (q). Now Sp2m (q) contains nontrivial scalar multiplications, which are fixed point free on U ∼ = Inn(E) in coprime action. Hence Hn (C(E), U ) = 0 for all n by (A2). So there is a faithful action of Sp2m (q) on E centralizing Z. Let X = E : Sp2m (q) be the corresponding semidirect product. Here the complement S ∼ = Sp2m (q) to E in X is determined up to conjugacy. X is a weak holomorph of E, and it is the unique one, up to isomorphism, as X = X 0 is perfect. X is a standard holomorph of E. For each faithful irreducible character θ of E is stable in X, and M(S) = 1 and X = X 0 imply that there is a unique χ ∈ Irr(X) extending θ. This forces that K(χ) = K(θ) = K. It remains to show that the automorphism α of E described above can be extended to X. Since it will permute then the nontrivial linear characters of Z transitively, and hence the faithful irreducible characters of E, this will prove that all faithful irreducible characters of X of degree q m are conjugate under automorphisms of X. Also we may alter it by an inner automorphism, if necessary, such that it fixes S. ¯ Using that X ¯ and S are perfect and Recall that A ∼ = X/Z = X. 1 ∗ ∗ ∼ M(S) = 1, and that H (S, U ) = 0 (U = U as S-modules) from (A2), ¯ Z) is the dual group of M(X) ¯ and that (A5) we infer that H2 (X, ¯ Z) → H2 (U, Z) Res : H2 (X, is injective. Since α fixes the conjugacy class of E in H2 (U, Z), being an automorphism of E, it therefore fixes the cohomology class of X in ¯ Z). This implies that α can be extended to X, as desired. We even H2 (X, b is the (universal) Schur cover of A, because the image of the see that X = A ¯ , and H2 (U, Z) ∼ restriction map is centralized by S ∼ = X/U = U ∗ ⊕ Λ2 (U )∗ as an S-module by (A8). We conclude that |M(A)| = |H2 (A, Z)| = |Z|, which gives the result. ¤

52


Theorem 4.3d. Let E = 21+2m . Up to isomorphism, there is a unique 0 standard holomorph X for E and its value field is K = Q(i). The faithful irreducible characters of X of degree 2m have Schur index 1 and are conjugate under automorphisms of X. Proof. Let Z = Z(E) and U = E/Z. Recall from Sec. 4.2 that Aut(E) = A : hαi where A = CAut(E) (Z) and where α is an involution inverting the elements of Z ∼ = Z4 and centralizing U ∼ = Inn(E) and C(E) = A/Inn(E) ∼ = Sp2m (2). Note that α is not an inner automorphism on A. Let θ ∈ Irr(E) be one of the two faithful irreducible characters of E. Then Q(θ) = K. Let first m ≥ 3. Then S = Sp2m (2) is a simple group, and the multiplier M(S) = 1 unless m = 3 (where it has order 2). In this situation it suffices to show that a weak holomorph for E exists, and a holomorph when m = 3. Embed E into GL2m (K) through θ (uniquely up to conjugacy). Let B be the block ideal in the group algebra KE to θ. This B is a centrally simple K-algebra isomorphic to M2m (K) which is stable under the action A. By the Skolem–Noether theorem there is a function τ : A → GL2m (K) such that conjugation with τx = τ (x) is application of x ∈ A. We have τxy = τx τy · τ (x, y) for x, y ∈ A, where τ (x, y) is a scalar matrix. Thus τ is a projective representation of A with 2-cocycle τ ∈ Z 2 (A, K ? ). We may arrange matters such that hτ (Inn(E))i ∼ = E. Let X0 be the subgroup of GL2m (K) generated by τ (A). Since the index |X0 : Z(X0 )| = |A| is finite, by a transfer argument due to Schur [Huppert, 1967, IV.2.3], X = X00 is finite. Then Z(X) consists of scalar matrices in K = Q(i) of finite order, that is, |Z(X)| is a divisor of 4. Since A = A0 is perfect and contains τ (E) ∼ = E, X is the desired holomorph. The case m = 1 is already treated above. For m = 2 we have S ∼ = S6 . 1+2 ∼ e e In this case consider E0 = E ◦ E, where E = E = 2 and where θ0 = θ ⊗ θe 0

is a faithful irreducible character of E0 . Let X0 be the standard holomorph e of E0 and χ0 ∈ Irr(X0 ) be the character extending θ0 . Let X = CX0 (E). e Then e is a subgroup of X0 on which χ0 decomposes as χ ⊗ θ. Then X ◦ E χ ∈ Irr(X) has its values in K. So X is a standard holomorph of E, and its uniqueness follows by noting that the other extensions 4.A6 .2 isoclinic to X require the 8th roots of unity. Also, the two extensions of θ to X are interchanged by a central automorphism of X. It remains to show that α can be extended to an automorphism of X. Let first m ≥ 4. Then H2 (S, Z) = 0 by (A5) since S = S 0 and M(S) = 1. Also, U ∗ = H1 (U, Z) is the dual module for S, and H1 (S, U ∗ ) = H1 (S, U )


53

has order 2 by (A9). From (A2) we infer that the kernel of the restriction map Res : H2 (A, Z) → H2 (U, Z) has order 1 or 2; its image is in the fixed subspace under S. Now the inclusion F2 ½ Z induces the zero map Ext(U, F2 ) → Ext(U, Z). It follows that the universal coefficient exact sequence splits naturally and so by (A8) yields the decomposition H2 (U, Z) ∼ = U ∗ ⊕ Λ2 (U )∗ of S-modules. As S is absolutely irreducible on U , HomS (U ⊗ U ∗ ) ∼ = EndS (U ) has order 2. Hence the commutator form to E in Λ2 (U )∗ = Hom(Λ2 (U ), Z) is the unique nontrivial element in H2 (U, Z) fixed by S. Consequently |M(A)| = |H2 (A, Z)| has order dividing |Z| = 4. However, b is the Z ½ X ³ A is a proper extension (Z ⊆ X 0 ). Hence X = X 0 = A universal Schur cover of A. By (A5) α can be lifted from A to X, and this lift cannot centralize Z as it is not inner on A. e where E e ∼ For m ≤ 3 consider E0 = E ◦ E = 21+2 . If X0 is the 0

standard holomorph of E0 and α0 a corresponding automorphism of X0 , e is the standard holomorph of E and α0 leaves E e and X then X = CX0 (E) invariant. Argue first for m = 3, then for m = 2, m = 1. ¤ Theorem 4.3e. Both E ∼ and E ∼ have two isomorphism = 21+2m = 21+2m + − √ types of standard holomorphs, with value fields Q( ±2) and Schur indices 1 or √ 2, the latter occurring when E is of negative type and the value field is Q( 2). The two standard holomorphs of either type (positive or negative) agree on the inverse images of Ω± 2m (2), and the faithful irreducible characm ters of degree 2 are rational-valued on these subgroups. These characters are always conjugate under group automorphisms. Proof. Let E be one of 21+2m or 21+2m . We may assume that m ≥ 2. + − √ Let θ be the unique faithful irreducible character of E. Let K = Q( −2). Observe that K is, in each case, a splitting field for E (as −1 is a sum of 2 squares in K). Arguing as in the proof for Theorem 4.3d we get a subgroup X0 of GL2m (K) for which Z(X0 ) consists of scalar matrices (including ±1) and X0 /Z(X0 ) ∼ = Aut(E), and we obtain that Y = X00 represents an 0 ∼ 0 extension of C(E) = O± 2m (2) by E. ± 0 ∼ ∼ 1+4 Now O± 2m (2) = Ω2m (2) is simple except when E = 2+ . At any 0 ∼ rate, let E0 = E ◦ E1 with E1 = D8 , and let Y0 = Y0 be the corresponding perfect subgroup of GL2m+1 (K). So Y0 /E ∼ = C(E0 )0 has index 2 in C(E0 ) ∼ = m+1 (K) interchanging two involutions in E1 O± 2(m+1) (2). There is x ∈ GL2

54


and centralizing E such that hE1 , xi is the semidihedral holomorph for D8 (which can be embedded into GL2 (K)). So x induces an orthogonal transvection on U0 = E0 /Z(E0 ), and x2 ∈ E1 . Since x is determined by this action on E0 up to multiplication with a scalar matrix, [Y0 , hxi] ⊆ Y0 and X0 = hY0 , xi is a holomorph for E0 . It follows that X = CX0 (E1 ) is a weak holomorph for E. The faithful irreducible character of X0 (given by its embedding into GL2m+1 (K)) remains absolutely irreducible on X ◦ E1 . We conclude that X is a holomorph of E. By construction there is χ ∈ Irr(X) extending θ which can be written √ in K = Q( −2). In particular Q(χ) ⊆ K. Let T be the inverse image in X X of Ω± 2m (2). We assert that ResT (χ) is rational-valued. This is obvious 0 when X = T , because θ is rational-valued and all extensions to X agree on X 0 . The case where E = 21+4 may be treated using that the elements of + Ω+ 4 (2) are just those which are products of an even number of transvections or reflections [Aschbacher, 1986, (22.14)], and appealing to Theorem 4.4 below. There are just two extensions of θ to X which agree on T , and these are interchanged by a central automorphism of X. e We claim that X is a standard holomorph of E. Write E = E0 ◦ E ∼ e e with E = D8 . Then X = CX (E0 ) is a (standard) holomorph of D8 . The e is irreducible and its value field is that of X. e Thus restriction of χ to E0 ◦ X √ e Q(χ) = K if X is a semidihedral group and Q(χ) ⊇ Q( 2) if it is dihedral. But the latter cannot happen as Q(χ) ⊆ K. However, replacing √ X by the ¤ isoclinic variant (with respect to T ) we obtain the value field Q( 2). 4.4. Good Conjugacy Classes Once Again We introduce a concept of “goodness” adapted to extraspecial groups. Sup1+2m pose X is a finite group containing some E ∼ as a normal subgroup. = q±,0 Let Z = Z(E) and U = E/Z. Following [Isaacs, 1973] an element x ∈ X is called “good for U” provided CU (x) = CE (x)/Z. In other words, whenever [x, y] ∈ Z(E) for some y ∈ E then [x, y] = 1. This depends only on the conjugacy class of Zx in X/Z, and also only on hZxi. Theorem 4.4. Let x ∈ X. There are exactly |U : CU (x)| cosets of Z in Ex which are good for U , and these are conjugate under E. Assume X has a faithful character χ which is absolutely irreducible on E. Then x is good for U if and only if χ(x) 6= 0, and then |χ(x)|2 = |CU (x)|. Furthermore, if x is a q 0 -element, then x is good for U , and χ(x) is a rational number provided χ is q-rational.


55

Proof. Let CU (x) = C/Z and D = CE (C). Then D/Z = (C/Z)⊥ is the orthogonal complement of CU (x) with respect to the (nondegenerate) symplectic commutator form on U . Hence |E : D| = |C : Z| and, therefore, |D : Z| = |E : C|. The group of automorphisms of C centralizing U = C/Z and Z is isomorphic to Hom(C/Z, Z) ∼ = C/Z ∼ = E/D, because C/Z is elementary and Z cyclic. Hence there is y ∈ E inducing the same central automorphism on C as it does x. It follows that xy −1 centralizes C. Clearly CU (xy −1 ) = CU (x). Hence xy −1 is good for U (and in the coset Ex). Now suppose x itself is good for U (for simplicity). Then all elements in the coset Dx are good for U . Conversely, to t ∈ E r D there exists c ∈ C such that [t, c] = z for some z 6= 1 in Z. It follows that [tx, c] = [t, c]x [x, c] = [t, c] = z, whence t is not good for U . Thus Dx is the set of elements in Ex which are good for U , and its cardinality is equal to |D| = |Z| · |E : C|. Now let χ ∈ Irr(X) be as indicated. Suppose χ(x) 6= 0. Without loss of generality assume that X = hE, xi. Then [x, y] = x−1 xy ∈ E for all y ∈ X. If [x, y] 6∈ Z then χ([x, y]) = 0 (as χ is an extension of a faithful irreducible character of E). If [x, y] = z for some z ∈ Z, then χ(x) = χ(xy ) = χ(xz) = χ(x)

χ(z) . χ(1)

We conclude that χ(z) = χ(1) and so [x, y] = z = 1, whence y ∈ CX (x). In particular x is good for U . (The argument even shows that x is good for Z or, equivalently, for the unique linear constituent of χ on Z.) Noting that χ is irreducible on X and applying Eq. (1.3c) we get |χ(x)|2 =

χ(1) X χ(1) χ([x, y]) = χ(1)|CX (x)|. |X| |X| y∈X

Now use that χ(1)2 = |U | (by the character theory of groups of extraspecial type) and that |X : CX (x)| = |E : CE (x)|. Suppose x is good for U . The set Dx consists of |E : C| distinct cosets mod Z, which are permuted by E via conjugation. But the stabilizer in E

56


of Zx is C. So these cosets are permuted transitively by E. By virtue of Eq. (1.3d) and using that χ vanishes on elements bad for U we therefore have X X X |E| = |χ(yx)|2 = |χ(yx)|2 = |E : C| |χ(zx)|2 . y∈E

y∈D

z∈Z 2

But for z ∈ Z we have |χ(zx)| = |χ(x)|. Thus |χ(x)| = |CU (x)|. Suppose x is a q 0 -element. Then CE (x)/Z = CU (x) by coprime action. Moreover, we have an orthogonal decomposition U = CU (x)⊥[U, x] with respect to the symplectic commutator form, and x acts faithfully and symplectically on [U, x] = [U, x, x]. Hence |CU (x)| = q 2n for some integer n, and if E0 is the inverse image in E of CU (x), then E0 is of extraspecial type (and CE (E0 ) maps onto [U, x]). It follows that |χ(x)| = q n , because x is good for U . By Theorem 1.6c χ(x) = ±µθ(1) for some sign and some root of unity µ of order dividing o(x). Hence if in addition χ is q-rational, then χ(x) is a rational number. ¤ 4.5. Some Weil Characters The classical groups GLm (q), Sp2m (q) (for odd q) and GUm (q) admit Weil characters. Usually these are faithful irreducible characters of smallest possible degree. The Weil characters of the symplectic groups were discovered by [Weil, 1964], [Ward, 1972] and [Isaacs, 1973]. Theorem 4.5a (Weil, Ward, Isaacs). Let S = Sp2m (q f ) for some power q f of the odd prime q and some m ≥ 1, and let U be its standard module. Then S has a pair of (disjoint) “generic” complex characters ξ 6= ξ ¦ , conjugate under an outer diagonal automorphism of S, such that ξ ξ¯ = πU (= ξ ¦ ξ ¦ ) is the permutation character on U . We have ξ = ξ1 + ξ2 where the ξi are irreducible characters of degree ξ1 (1) = (q f m −1)/2 and ξ2 (1) = (q f m +1)/2 (and similar statement for ξ ¦ ). The following hold: ¢ ¡p (i) The field of character values for ξ, ξ1 and ξ2 is Q (−1)(q−1)/2 q if f is odd, and Q otherwise. The characters take only rational values on q 0 -elements of S. (ii) Just one of ξ1 , ξ2 is faithful for S, and ξ1 is faithful if and only if q f m ≡ 1 (mod 4). Moreover, ξ1 is irreducible as a Brauer character in any characteristic different from q, ξ2 in every characteristic different from q and from 2, and ξ2 = 1S + ξ1 on 20 -elements of S.


57

Proof. The trace map Fqf → Fq carries the symplectic Fqf -form on U to one over Fq which remains nondegenerate and S-invariant. This yields an embedding of S = Sp2m (q f ) into Sp2f m (q), hence an embedding of S into 1+2f m the standard holomorph of E ∼ (Theorem 4.3c). Let Z = Z(E) and = q+ identify U = E/Z (as S-modules). The above discussion carries over to the symplectic holomorph Xs = E : S (contained in the standard holomorph of E). Recall the definition of the automorphism α of E in Sec. 4.2, via a decomposition U = W ⊕ W ∗ into totally isotropic subspaces. As before we may extend this α to X and pick S to be α-invariant. Then α induces the outer (diagonal) automorphism of order 2 on S and permutes the q − 1 nontrivial linear characters of Z transitively. The restrictions to S of (the) q − 1 faithful irreducible characters of X of degree q f m lying above these linear characters give rise to two distinct generic Weil characters ξ, ξ ¦ of S conjugate under α. For (m, q f ) = (1, 3) we have additional central automorphisms, and here we we pick ξ, ξ ¦ such that they do not contain the 1-character of S (cf. Example (iv) in 4.3b). Let j be the central involution in S. Let x ∈ S. We claim that x is good for U . Let t ∈ E such that Zt ∈ CU (x). Since j inverts the elements in U , we have tj = t−1 z for some z ∈ Z and [x, t] = [x, t]j = [x, tj ] = [x, t−1 z] = [x, t]−1 . Using that q is odd this implies that [x, t] = 1. Hence the claim. It follows ¯ that (ξ ξ)(x) = |ξ(x)|2 = |CU (x)| by Theorem 4.4. Thus ξ ξ¯ = πU is the permutation character of S on U . In particular |ξ(j)|2 = |CU (j)| = 1 and so ξ(j) = ±1 as it is a rational number. Since S is transitive on U ] (by Witt’s theorem), it follows that ¯ 1S i = 2. hξ, ξi = hξ ξ, Consequently ξ = ξ1 + ξ2 for two distinct irreducible characters ξi of S. So ξi (j) = ±ξi (1) for i = 1, 2. It follows that just one of the characters ξi has j in its kernel, and that ξi (1) = (q f m ± 1)/2. Choose notation such that ξ2 (1) = ξ1 (1) + 1. Let x ∈ S be a symplectic transvection. Then |ξ(x)|2 = |CU (x)| = q . Since S is generated by the symplectic transvections, this gives the result on the character field of ξ, up to the sign. For the precise sign we refer to [Isaacs, 1973] (Theorem 5.7), from which one also infers that ξ(j) = 1 if and only if q f m ≡ 1 (mod 4) (see also [I, 13.32]). Use further f (2m−1)

58


that 1S + ξ1 and ξ2 agree on q-elements (as proved in the next paragraph), and that ξ1 is faithful if and only if ξ1 (j) = −ξ1 (1), that is, if and only if ξ(j) = +1. (We mention that ξ ¦ = ξ¯ if and only if f is odd.) Let us pass to characteristic p for some prime p 6= q. Let F be the residue class field of Z(p) [e2πi/q ] modulo some maximal ideal, and let V be an F S-module affording ξ as a Brauer character. Then, as before, V ⊗F V ∗ is the permutation module of S over F on the set U (with two orbits). Thus ∼ HomF S (V ⊗F V ∗ , F ) EndF S (V ) = has F -dimension 2. If p 6= 2 then V = [V, j] ⊕ CV (j) is a proper decomposition into F S-modules, because j is in the kernel of just one ξi . So let p = 2. Then j acts trivially on each irreducible F S-module, but it is nontrivial on V as it is nontrivial on U . Hence CV (j) is a proper submodule of V and V is not completely reducible. As before V /CV (j) ∼ = [V, j] as F S-modules (via the commutator map v 7→ [v, j]). We cannot have CV (j) = [V, j] as q is odd. We cannot have CV (j) ⊂ [V, j] (properly), because then CV (j) were irreducible and [V, j] = CV (j) ⊕ 1S , which is impossible. We conclude that [V, j] is irreducible and of codimension 1 in CV (j), the quotient being the trivial module (affording 1S ). The proof is complete. ¤ The characters ξ, ξ1 , ξ2 of Sp2m (q f ) obtained in Theorem 4.5a, and their conjugates, are called the Weil characters of the symplectic group. For m ≥ 2 the irreducible Weil characters are the unique faithful irreducible characters of Sp2m (q f ) of degree less than q f m − q (see Appendix C). , m ≥ 3, Theorem 4.5b. Let X be a standard holomorph of E ∼ = 21+2m + and let χ be one (of the two) faithful irreducible characters of X of degree 2m . Write E as a central product of m dihedral groups hai , a∗i i of order 8, with involutions ai , a∗i . Let A = ha1 , · · ·, am i and A∗ = ha∗1 , · · ·, a∗m i. ∼ Lm (2) of X normalizing both (i) There exists a unique subgroup L = A and A∗ , so that A is the standard module for L and A∗ its dual. We have NX (L) = ZLhτ i where τ 2 ∈ Z = Z(E) and where τ interchanges A and A∗ and induces the inverse transpose automorphism on L. (ii) ResX L (χ) = πA = 2 · 1L + ξ where ξ is an (absolutely) irreducible (Weil) character of L which is rational-valued (with Schur index 1). This ξ is irreducible modulo every odd prime p not dividing 2m − 1, and reducible otherwise. Proof. We use the Atlas notation for the projective special linear groups. So at present Lm (2) = PSLm (2). Let Z = Z(E) and U = E/Z.


59

∼ O+ (2). Let W = AZ/Z and W ∗ = ¯ = X/E = (i) Recall that X 2m A∗ Z/Z. Then U = W ⊕ W ∗ is a decomposition into maximal totally singular subspaces. Let P = NX¯ (W ), a maximal parabolic subgroup of ∗ ¯ Ω+ 2m (2) (see Appendix B). Let L = NP (W, W ) be a Levi complement, so ∼ ¯ = Lm (2) and W is the standard L-module, ¯ that L W ∗ its dual. See Appendix (C1) for the existence (and uniqeness) of the subgroup L of X. This L is faithful on A and A∗ , and maps isomorphically onto ¯ For each i let ei = ai Z and e∗ = a∗ Z, and let τ¯i be the orthogonal L. i i transvection of U with centre ei · e∗i . Then τ¯ = τ¯1 · · · τ¯m is an involution in ¯ interchanging each ei , e∗ , hence W and W ∗ . Also, τ¯ normalizes L ¯ and X i induces on it the inverse transpose automorphism. ¯ The L-modules W and W ∗ are not isomorphic (Lemma 4.1c). Hence ¯ permutes {W, W ∗ }. Since L ¯ = NX¯ (W, W ∗ ), we see that NX¯ (L) ¯ = NX¯ (L) ¯ Aut( L) ¯ τ i = Aut(L). ¯ We observe that U = Ind ¯ ¯ Lh¯ (W ) as an Aut(L)-module and so

L

¯ U) ∼ ¯ W) ∼ ¯ W ∗) Hn (Aut(L), = Hn (L, = Hn (L,

by Shapiro’s lemma (A3). This vanishes for n = 1, 2 when m ≥ 6 by (A9). Then we find τ ∈ X mapping onto τ¯ with τ 2 ∈ Z, and replacing τ by a suitable E-conjugate, if necessary, this τ normalizes L. Note that NE (L) = Z as L has no fixed points on U = E/Z. It follows that NX (L) = ¯ For 3 ≤ m ≤ 5 we argue as follows. Lhτ iNE (L) = Lhτ iZ maps onto Aut(L). 1+2(6−m) ∼ . Let X0 be the standard Consider E0 = E ◦ E1 , with E1 = 2+ holomorph of E0 having the same value field as X. Then X = CX0 (E1 ) (by uniqueness). Let L0 ∼ = L6 (2) be the subgroup of X0 normalizing A0 = ha1 , ·· ∗ ∗ ·, a6 i and A0 = ha1 , ···, a∗6 i. Then L = CL0 (E1 ), because this centralizer is a subgroup of X isomorphic to Lm (2) which normalizes A and A∗ . For each i T let τi ∈ j6=i CX0 (haj , a∗j i) be an element of the holomorph of hai , a∗i i within X0 mapping onto τ¯i . Let τ0 = τ1 · · · τ6 . Then L0 Z and Lτ00 Z are conjugate in EL0 = (EL0 )τ0 (as H1 (L0 , U0 ) = 0 for U0 = E0 /Z(E0 )). Hence there are yi ∈ hai , a∗i i such that for y0 = y1 ···y6 we have (L0 Z)τ0 y0 = L0 Z, hence Lτ00 y0 = (Lτ00 y0 Z)0 = L0 . Also τ0 y0 normalizes E1 and hence L = CL0 (E1 ). Now τ = τ1 · · · τm y1 · · · ym maps onto τ¯ and centralizes E1 , hence is in X = CX0 (E1 ). Using that τi permutes with yi for j 6= i and that L is centralized by τi , yi when i > m we see that τ normalizes L. (ii) It follows from Appendix (C1) that ResX L (χ) = πW = 2 · 1L + ξ for 0 some irreducible character ξ of L. Since L = L ⊆ X 0 , the Weil character ξ of L is rational-valued and independent of the choice of χ.

60


Let p be an odd prime. We investigate the (transitive) permutation character πW ] = 1L +ξ, and its reduction mod p. Let M be the permutation Fp L-module affording πW ] (as a Brauer character). If p - 2m − 1 then M = M0 ⊕ 1L where M0 is (absolutely) irreducible (affording ξ as a Brauer character). So let p | 2m − 1. Then 1L appears at least twice in M , namely f of M , the in the socle and in the head. (One can show that the “heart” M section remaining, is irreducible (and not trivial): Since L = L0 is perfect, f cannot have 1L in the socle or in the head. By [Seitz–Zalesskii, 1993] M f either is trivial or has dimension ≥ 2m −m−1. each composition factor of M m e Hence M f is irreducible.) ¤ We have 2 · (2 − m − 1) > 2m − 3 = dim M. Remark . Replace E ∼ by E0 = E ◦ Z4 ∼ (m ≥ 3). There = 21+2m = 21+2m + 0 is a corresponding result for the standard holomorph X0 of E0 . More precisely, let Y = NX0 (E). Then Y /E0 ∼ = O+ 2m (2) is a maximal subgroup of ¯ 0 = X0 /E0 ∼ X Sp (2). As before we find L∼ = 2m = GLm (2) in Y normalizing ∗ ¯ ¯ ¯ τ i as before, and we find A and A . For L = LE0 /E0 we have NX¯ 0 (L) = Lh¯ τ ∈ X0 mapping onto τ¯, normalizing L and interchanging A, A∗ such that τ 2 ∈ Z(E0 ) and NX0 (L) = Lhτ iZ(E0 ). 4.6. Symplectic and Orthogonal Modules Now we state and prove two results, due to [Isaacs, 1973] and [Gow, 1993], which will be crucial for our approach to the k(GV ) theorem. Theorem 4.6a (Isaacs). Let V is a coprime symplectic Fp G-module. There exists a rational-valued character χ of G with χ2 = πV , and this yields a rational-valued generalized character ψ of G with ψ(1) = 1 and ψ 2 = δV . Proof. We may assume that G is faithful on V . Let |V | = p2m . In the odd case let χ be the restriction to G of a generic Weil character of Sp2m (p) (Theorem 4.5a). Otherwise identify V = E/Z(E) via the commutator form on E = 21+2m , and let T be the standard holomorph of E (Theorem 4.3d). 0 By the Schur–Zassenhaus theorem we may embed G into T . Let then χ be the restriction to G of one of the (two) faithful irreducible characters of T of degree 2m . Apply Theorem 4.4. Let X = GV be the semidirect product, and view χ as a character of X e is a generalized character of X, because by inflation. By Lemma 2.1a, p1m χ e(x)/|CV (x)| = χ(1) = pm and |X|p = |V | = p2m . Define ψ(x) = p1m χ m p χ(x)/|CV (x)| for x ∈ G. By Lemma 3.3a, ψ is a generalized character of G. We have ψ(1) = 1 and ψ 2 = δV , and ψ is rational-valued. ¤


61

Theorem 4.6b (Gow). Suppose V is a coprime orthogonal Fp G-module for some odd prime p. Let N = Ker(µ) be the kernel of the linear character µ of G afforded by the determinant in its action on V (|G/N | = 1 or 2). There is a rational-valued generalized character ψ of G such that ψ(1) = 1 and ψ 2 = δV on N , ψ(x)2 = p1 δV (x) for x ∈ G r N (if any). By virtue of Lemmas 2.1a and 3.3a it suffices to construct a rational-valued generalized character χ of G with the following properties: If dim V = 2m is even, then χ(1) = pm , χ2 = πV on N and χ(x)2 = p1 |CV (x)| for x ∈ G r N . If dim V = 2m + 1 is odd, then χ(1) = pm+1 , χ(x)2 = p|CV (x)| for x ∈ N and χ(x)2 = |CV (x)| otherwise. We need two lemmas. Let τ = [·, ·] be the G-invariant, nondegenerate symmetric bilinear form on V . If dim V = 2m is even, we write ε(V ) = 1 if τ has Witt index m and ε(V ) = −1 otherwise. Similar notation for nondegenerate subspaces of even dimension (ε(0) = 1 for the zero subspace). We set d(V ) = 1 if the discriminant of τ is a square in F?p and d(V ) = −1 m otherwise. Thus ε(V )d(V ) = (−1)(p −1)/2 if dim V = 2m. In the odd m dimensional case dim V = 2m + 1, we define ε(V ) = (−1)(p −1)/2 d(V ), and extend this definition to nondegenerate subspaces of odd dimension. Lemma 4.6c. Let I(V ) be the set of vector v ∈ V for which τ (v, v) = 0, and let J(V ) consist of those with τ (v, v) = 1. (i) If dim V = 2m, then |I(V )| = p2m−1 + ε(V )(pm − pm−1 ) and |J(V )| = p2m−1 − ε(V )pm−1 . (ii) If dim V = 2m + 1, then |I(V )| = p2m and |J(V )| = p2m + ε(V )pm . Proof. Recall that V = U1 ⊥ · · · ⊥Um−1 ⊥V0 with hyperbolic planes Ui = hui , vi i, satisfying τ (ui , ui ) = 0 = τ (vi , vi ) and τ (ui , vi ) = 1, where either V0 = Um is a hyperbolic plane (dim V = 2m and ε(V ) = 1), or V0 = hum , vm i with τ (um , vm ) = 0, τ (um , um ) = 1 and τ (vm , vm ) = −c for some nonsquare c in Fp? (dim V = 2m and ε(V ) = −1), or V0 = hvm i where either τ (vm , vm ) = 1 or τ (vm , vm ) = c is a nonsquare (dim V = 2m + 1). It is easy to verify the assertions when dim V ≤ 2. Observe that if U is a hyperbolic plane, to every isotropic u 6= 0 in U there exists a unique isotropic v 6= 0 in U such that {u, v} is a hyperbolic pair, and then the set of vectors au, av for a ∈ Fp are just all isotropic ones in U (including 0). Hence |I(U )| = 2p − 1 and |J(U )| = p − 1.

62


So let dim V ≥ 3, and write V = U ⊥W where U is a hyperbolic plane. One checks the following recursion formulas (dim W = k = dim V − 2): |I(V )| = pk+1 − pk + p|I(W )|, |J(V )| = (2p − 1)|J(W )| + (p − 1)(pk − |J(W )|). The lemma follows.

¤

Clearly I = I(V ) and J = J(V ) are stable under the action of G. We consider the permutation characters πI and πJ of G on these sets, and define χ = ε(V )(πI − πJ ) if dim V = 2m is even. Then χ(1) = pm . Let dim V = 2m + 1. Then we p−1 0 0 let χ0 = −ε(V )(πI − πJ ) and define χ by χ = p+1 2 χ + 2 χ · µ, that is, χ = pχ0 on N and χ(x) = χ0 (x) for x ∈ G r N . Then χ(1) = pm+1 . The computation of this rational-valued generalized character χ of G will prove that it has the asserted properties, hence will complete the proof of Theorem 4.6b. Let x ∈ G. Since G is a p0 -group, we have an orthogonal decomposition V = [V, x]⊥CV (x) into nondegenerate subspaces. If det(x) = 1 then [V, x] has even dimension, whence dim V and dim CV (x) the same parity, whereas dim [V, x] is odd if det(x) = −1. Let dim CV (x) = rx and εx = ε([V, x]). rx

Lemma 4.6d. (i) Let dim V = 2m. Then χ(x) = ±εx pb 2 + holds if det(x) = 1 or det(x) = −1 and p ≡ 3 (mod 4). (ii) Let dim V = 2m + 1. Then χ(x) = ±εx pb holds if det(x) = 1.

rx +1 c 2

c

where the sign

where the + sign

Proof. Suppose first that det(x) = 1. If dim V = 2m, then rx is even and by Lemma 4.6c, the very definition of χ, and elementary properties of the Witt index, we get χ(x) = ε(V )ε(CV (x))prx /2 = εx prx /2 . If dim V = 2m + 1, then rx is odd, and we obtain χ(x) = ε(V )ε(CV (x))p(rx +1)/2 = εx p(rx +1)/2 . Suppose next that det(x) = −1. Then rx is even precisely when dim V is odd (and then b rx2+1 c = r2x ). We obtain that χ(x) = −ε(V )ε(CV (x))pb

rx 2

c

.

It is elementary to show that ε(V )ε(CV (x)) = εx if dim V = 2m + 1 and ε(V )ε(CV (x)) = (−1)(p−1)/2 εx if dim V = 2m. We are done. ¤

Chapter 5

Real Vectors The Robinson–Thompson theorem shows that k(GV ) ≤ |V | provided there is a real vector in V for G. This is fundamental for our approach to the k(GV ) problem. Here a vector v ∈ V is called real for G if the restriction to CG (V ) of V contains a faithful self-dual submodule (with a real-valued Brauer character). The search for real vectors is compatible with Clifford reduction and leads to the nonreal reduced pairs (which will be classified in the next two chapters). 5.1. Regular, Abelian and Real Vectors Throughout V is a coprime F G-module, not necessarily faithful, where F = Fr is a finite field of characteristic p. Let v ∈ V and let H = CG (v). The vector v is called regular for G provided H = CG (V ), and it is called abelian (cyclic) if H/CG (V ) is abelian (cyclic). If G is faithful on V , then k(GV ) ≤ |V | by Theorem 3.4d if there is an abelian vector in V for G. We say that v is strongly real for G provided the restriction to H of V is self-dual, and it is called real if ResG H (V ) contains a self-dual submodule W such that CH (W ) = CG (V ). Of course regular vectors are abelian and are strongly real, whereas there is no hierarchy between abelian and (strongly) real vectors. There are examples (G, V ) where there are neither abelian vectors nor real ones, or vectors of just one kind (see for instance Secs. 6.1, 7.1 below). Example 5.1a. We are going to describe the permutation pairs (G, V ). Let p > d + 1 and let E = Ad+1 be the alternating group of degree d + 1 ≥ 5. Then the deleted (shortened) permutation F E-module V of degree d is coprime, faithful and absolutely irreducible. It is the “heart” of the Ld permutation module W = {wi } is a permutation basis i=0 F wi , where P P (wi s = wis for s ∈ E). It consists of all i ci wi with i ci = 0 in F (as p - d + 1). Embed E into GL(V ), and let G = NGL(V ) (E). Then G = S × Z where Z = Z(G) ∼ = F ? , and where we may choose S ∼ = Sd+1 such that it acts on V as its natural shortened permutation module. (For d + 1 = 6 note that PGL2 (9) and M10 do not have faithful actions on 5-dimensional coprime modules.) There exist vectors of (almost) every kind here. 63

64


(i) There is always a strongly real vector v ∈ V for G: Pd Let v = dw0 − i=1 wi . If g = xz fixes v (x ∈ S, z ∈ Z), then zwix = wj for some positive i, j, hence z = 1, and g = x must fix w0 and permute {w0 − wi }di=1 . Therefore H = CG (v) ∼ = Sd and ResG H (V ) is the natural permutation over F , which gives the result. (ii) There is always an abelian (cyclic) vector in V for G: Pd A vector u = W is regular for S if and only if all ci ∈ F i=0 ci wi in Q d are distinct. So there are i=0 (r − i) regular vectors in W for S. If P j such that cj − c 6= ci for all i 6= j, i ci = c (6= 0), there is a (first) index Qd because p > d + 1. Hence there are i=1 (r − i) vectors in V regular for S. If u is regular for S, then CG (u) is cyclic of order dividing |Z| = r − 1. (iii) There is a regular vector in V for G if and only if r ≥ d + 4: ¡r−1¢ Qd 1 1 regular S-orbits We have seen that there are (d+1)! i=1 (r − i) = d+1 d orbits for on V ] . So we have one such orbit when r = d + 2, and r−1 2 r = d + 3. In these two cases (where r = p) there cannot be a regular G-orbit on V ] , because each regular G-orbit on V ] is a disjoint union of |Z| = r − 1 regular S-orbits. Let r ≥ d + 4 in what follows. Suppose g = xz (x ∈ S, z ∈ Z) is an element in G of prime order s P which fixes a vector u = i ci wi in V belonging to a regular S-orbit. Then cix = ci z for all i, hence jx = j if some (unique) cj = 0, and the other cycles of x have the same size s = o(x) = o(z), with cix = ci z, · · ·, cixs = ci z s = ci . Hence s is a divisor of r − 1 and of either d or d + 1, and z is determined by x. Let δ = δs ∈ {0, 1} be (unique) such that s | d + δ. We see that g Q d+δ s −1 (r − 1 − is) vectors in V regular for S. There are just fixes at most i=0 (d+1)! elements in S = Sd+1 of the cycle shape of x [James–Kerber, d+δ d+δ s

s

(

s

)!

1981, 1.2.15]. Using that each group ¡(r−1)/sof¢ order s has s − 1 generators we get that there are at most (d+1)! s−1 (d+δ)/s vectors in V belonging to regular S-orbits which are fixed by subgroups of order s in G. Summing up over the primes s dividing both d(d + 1), i.e., d + δs for some δs ∈ {0, 1}, and r − 1 one obtains that (d + 1)!

X s

µ ¶ Y d (r − 1)/s 1 < (r − i). s − 1 (d + δs )/s i=1

This implies that there is a regular vector in V for G, because there is a vector regular for S which is not fixed by any nontrivial element of G.

Real Vectors

65

Confirm the inequality first for r = d + 4, where r = p is odd, d is even and only s = 2, 3 can appear (the latter when 3 | r − 1); use that r ≥ 7 d+δ2 r−1 s − d+δ (even r ≥ 11). For any s we have r−1 2 − s ≥ 2 s . Therefore ¡ (r−1)/s ¢ ¡ (r−1)/2 ¢ ≤ (d+δ2 )/2 , blowing up the (d + δs )/s-subsets of an (r − 1)/s-set (d+δs )/s to (d ¡+ δ2 )/2-sets in an underlying (r − 1)/2-set. Hence ¡r−1¢ it suffices to show ¢ Qd (r−1)/2 P 1 1 1 that (d+δ)/2 i=1 (r − i) = d+1 s s−1 < (d+1)! d . Blowing up the (d + δ)/2-subsets of an (r − 1)/2-set to d-sets in an underlying (r − 1)-set ¡ ¢ ¡(r−1)/2¢ we get that r−1 > α where d (d+δ)/2 µ ¶ µ ¶ r − 1 − (r − 1)/2 (r − 1)/2 α= ≥ = (r − 1)(r − 3)/8. d − (d + δ)/2 2 Here (r − 1)/2 − [d − (d + δ)/2] ≥ 2 as we assume that d + 1 < r − 3. It is thus enough to check that X 1 ≤ (r − 1)/8, s−1 s with s ranging over the primes dividing r − 1 (say). This is immediate for r = 11, 13. If s is an odd prime divisor of r − 1, then 2s | r − 1 and P P P 1 1 1 1 + 1 . Thus s s−1 ≤ t|r−1 1t = r−1 t|r−1 t, which is known to s−1 ≤ ¡ s 2sε ¢ P be O (r − 1) for each ε > 0. For r ≥ 17 we have t|r−1 t ≤ (r − 1)2 /8. Lemma 5.1b. Let G be faithful on V , and assume that G = X ◦ Z is a central product over Z(X) where Z acts on V as a group of scalar multiplications. Let H = CX (v) for some v ∈ V ] . If |NX (H) : H| is relatively prime to |Z : Z(X)|, or if NX (Zv) = H × Z(X), then CG (v) = H. Hence if v is (strongly) real (or abelian, regular) for X, then it is so for G. Proof. We write Zv = vZ regarding the elements of Z (acting) as scalars (modules being right modules). Clearly N = NX (Zv) is contained in NX (H), and CG (v) ⊆ NG (Zv) = N Z. To any y ∈ N there exists a unique z = zy in Z such that yz ∈ CG (v). Conversely, each x ∈ CG (v) may be written as x = yz with y ∈ X and z ∈ Z, and then y ∈ N and z = zy . The assignment y 7→ zy is a homomorphism from N to Z with kernel H = CX (v). Under either assumption the image of this homomorphism is in Z(X), whence CG (v) ⊆ N Z(X) ⊆ X. ¤ Remark . We have CG (v) ∩ Z = 1 and CG (v)Z/Z = CG/Z (Zv) for each nonzero vector v ∈ V . In most situations Z ∼ = F ? will be the group of all ] scalar multiplications. Then Zv = (F v) and CG/Z (F v) = CG/Z (Zv) is nothing but the stabilizer of the point F v in the projective 1-space P1 (V ).

66


5.2. The Robinson–Thompson Theorem We begin by combining Theorems 4.1b, 4.6a and 4.6b. Lemma 5.2a. Suppose V is a coprime Fp G-module and v ∈ V is real for G. Let H = CG (v) and let U be a self-dual Fp H-submodule of V with CH (U ) = CG (V ). Then there is a rational-valued generalized character ψ of H and a subgroup N of H with |H : N | ≤ 2 such that ψ(1) = 1, ψ 2 = δU on N and ψ(h)2 = p1 δU (h) whenever h ∈ H r N . We have N 6= H if and only if p is odd and some element of H acts with determinant −1 on U . Proof. According to Theorem 4.1b we may write U = U0 ⊕ W0 where U0 is a symplectic Fp H-module and W0 is an orthogonal module, the action of H on W0 being trivial when p = 2. By Theorem 4.6a there is a rational-valued 2 generalized character ψU0 of H satisfying ψU0 (1) = 1 and ψU = δU0 . If 0 H acts trivially on W0 , then we let N = H and ψW0 = 1H . Otherwise p 6= 2, and we let N be the kernel of the linear character of H afforded by the determinantal action of H on W0 (and on U ). By Theorem 4.6b there is a rational-valued generalized character ψW0 of H satisfying ψW0 (1) = 1, 2 ψW = δW0 on N , and ψW0 (h)2 = p1 δW0 (h) for h ∈ H r N . Now define 0 ψ = ψ U 0 ψ W0 . ¤ Theorem 5.2b (Robinson–Thompson). Suppose V is a faithful coprime Fp G-module and some v ∈ V is real for G. Then k(GV ) ≤ |V | and equality can only hold when v is strongly real for G. Also, k(GV ) < |V | if p ≥ 5 and some element of H acts with determinant −1 on V . Proof. Let H = CG (v), and let U , ψ be as in the preceding lemma. Since V is faithful by hypothesis, U is a faithful (self-dual) Fp H-submodule of V here. Thus U and ψ are as assumed in Theorem 3.3d. It follows that k(GV ) ≤ |V and that equality only holds when [V, H] ⊆ U . But in this case V = U ⊕ U 0 as an Fp H-module with H acting trivially on U 0 . Thus V itself then is self-dual for H. Suppose p is odd and some element of H acts with determinant −1 on V . Assume k(GV ) = |V |. Then V is self-dual for H, as seen above. So Lemma 5.2a and Theorem 3.3d apply. We conclude that k(GV ) ≤ p+3 2p |V |, and this is less than |V | when p ≥ 5. ¤ In [Robinson–Thompson, 1996] the conditions under which the assumptions in Theorem 3.3d are fulfilled have been carefully studied. Let

Real Vectors

67

us describe this briefly. Assume (without loss of generality) that V is irreducible (Proposition 3.1a), and faithful, and let χ ˘ be the Brauer character of G afforded by V . So, as in Sec. 2.7, χ ˘ = TrKD (χ)|KD (χ) for some ordinary absolutely irreducible character χ of G, where KD is the decomposition field of p in K = Q(e2πi/e ) for e = exp (G). This χ is the Brauer character of G afforded by some absolutely irreducible constituent of V , and it is faithful as algebraically conjugate characters have the same kernel. Let Γ = Gal(K|Q) and ΓD = Gal(K|KD ). Let H = CG (v) for some v ∈ V . If θ is an irreducible constituent of ResG (χ), then θ˘ = TrK (θ)|K (θ) is H

D

D

a Brauer character of H afforded by some irreducible constituent W = Wθ ¯ of ResG H (V ). By Lemma 4.1a, W is self-dual if and only if θ and θ are Galois conjugate over KD . (Use that Γ is abelian.) If τ ∈ Γ r ΓD , then W τ = Wθτ affords TrKD (θ)|KD (θτ ) and is an irreducible Fp H-module not isomorphic to W . But δW τ = δW (see Sec. 2.7). Moreover, W τ is self-dual if and only if W is self-dual. Now consider first all irreducible constituents θ of ResG H (χ) for which hχ, θiH ≥ 2. Then ResG (V ) contains a submodule U = Wθ ⊕ Wθ , and θ H 2 τ δUθ = δW . Consider next those θ for which both θ and θ are irreducible θ G constituents of ResG (χ) for some τ ∈ Γ r Γ . Then Res (V ) contains a D H H τ 2 submodule Uθ,τ = Wθ ⊕ Wθ , and again δUθ,τ = δWθ . Let U0 be the sum of all these submodules Uθ , Uθ,τ of ResG H (V ). Then there is a rational-valued generalized (Knörr) character ψ0 of H such that ψ02 = δU0 . If U0 is a faithful Fp H-module, we are done. Otherwise we may search ¯ for irreducible constituents θ of ResG H (χ) for which θ and θ are conjugate over KD . In this case Uθ = Wθ is self-dual, and Theorems 4.1b, 4.6a and 4.6b apply. If we add all these modules to U0 , the resulting Fp H-module U is a submodule of ResG H (V ), and it can be faithful, or not. Proposition 5.2c. Let V be a coprime, faithful, irreducible Fp G-module affording TrKD (χ)|KD (χ) for some (absolutely) irreducible character χ of G. Let H = CG (v) for some v in V . Suppose there are irreducible constituents θi of ResG H (χ) which are pairwise not algebraically conjugate and satisfy one of the following: hθi , χiH ≥ 2, hθiτ , χiH ≥ 1 for some τ ∈ Γ r ΓD , P or θ¯i is conjugate to θi over KD . If then i θi is a faithful character of H, the assumptions in Theorem 3.3d are fulfilled (for appropriate U and ψ determined by the θi ). Proof. Clear in view of the above discussion.

¤

68


5.3. Search for Real Vectors Let now V be an arbitrary coprime F G-module, not necessarily faithful. It is obvious that if the direct summands of V admit (strongly) real vectors, then so does the module itself. Our objective is to ensure that the search for (strongly) real vectors is compatible with Clifford reduction. Proposition 5.3a.. Let V = IndG H (U ) be induced from the F H-module U . (i) If there is a (strongly) real vector in U for H, there is a (strongly) real vector in V for G. (ii) If there is a regular vector in V for N = CoreG (H), there is a real vector in V for G. T Proof. Let {ti } be a right transversal to H in G (so that N = i H ti ). P (i) Let u ∈ U be real for H, and let v = i uti . Let W be a selfdual F CH (u)-submodule of U with CH (W ) ∩ CH (u) = CH (U ), and let L V0 = i W ti . Then V0 is a subspace of V . Now CG (v) ∩ H ti centralizes uti and so W ti is self-dual as an F [CG (v) ∩ H ti ]-module. Inducing up this selfdual module to CG (v) yields a self-dual F CG (v)-module by Lemma 4.1a. By Mackey decomposition (1.2d) V0 is a direct sum of such modules (taken over a set of representatives for the orbits of CG (v) on the cosets Hti ). It remains to show that CG (V0 ) ∩ CG (v) = CG (V ). Now the centralizer in CG (v) of V0 preserves each coset Hti and so lies in N . It follows that CG (V0 ) ∩ CG (v) ⊆ N ∩ CH (W ) ∩ CH (u) = CN (U ). Similarly CG (V0 ) ∩ CG (v) ⊆ CN (U ti ) for T each i, and i CN (U ti ) = CG (V ). The statement for strongly real vectors is treated similarly. P (ii) Let v = i uti be a regular vector in V for N . We may assume that all vi = uti 6= 0. Let Htj CG (v) be the different double cosets of G modulo (H, CG (v)). By Mackey decomposition once again ResG CG (v) (V ) =

M j

Let V0 =

L j

C (v)

IndCG (v)∩H tj (U tj ). G

C (v)

IndCG (v)∩H tj (F vj ). Then V0 is a permutation module for G

CG (v) over F , hence self-dual. Since each F vj has a CG (v) ∩ H tj -invariant complement in U tj (Maschke), and since module induction respects direct sums, V0 is a direct summand of V . As before CG (V0 ) ∩ CG (v) ⊆ N . Hence CG (V0 ) ∩ CG (v) ⊆ N ∩ CG (v) = CN (v) = CG (V ), as desired. ¤

Real Vectors

69

Proposition 5.3b. Suppose V = U ⊗F W for some (coprime) F G-modules U , W with 2 ≤ d = dim F U ≤ dim F W (say). (i) If there are (strongly) real vectors in U and in W for G, there is a (strongly) real vector in V for G. Similar statement for regular vectors. (ii) Let {ui }, {wj } be F -bases of U and W , respectively. Let v = i=1 ui ⊗ wi , and let W0 be the subspace of W generated by w1 , · · ·, wd . Then V0 = U ⊗F W0 is a self-dual F CG (v)-module. f of W such (iii) Assume that there is a (d − 1)-dimensional subspace W f that CG (W )/CG (W ) has a regular orbit on W . If G induces all scalar Pd

multiplications on W , there is a real vector in V for G. Proof. (i) We have CG (U )∩CG (W ) ⊆ CG (V ), the elements in CG (V ) being those elements of G inducing on U and W scalar multiplications which are inverse to each other. For if u ∈ U , w ∈ W are nonzero vectors, v = u ⊗ w and g ∈ CG (v), then v = vg = ug⊗wg implies that ug = cu and wg = c−1 w for some unique scalar c = cg (v) ∈ F ? . If u, w are regular vectors for G, that is, CG (u) = CG (U ) and CG (w) = CG (W ), then CG (v) = CG (V ). Let u ∈ U and w ∈ W be real vectors for G. Let U0 and W0 be self-dual submodules of U and W for CG (u) and CG (w), respectively, with CG (u) ∩ CG (U0 ) = CG (U ) and CG (w) ∩ CG (W0 ) = CG (W ). We may and do assume that u ∈ U0 and w ∈ W0 . (Otherwise replace U0 by U0 ⊕ F u; W0 by W0 ⊕ F w.) Let v = u ⊗ w and V0 = U0 ⊗F W0 . If g ∈ CG (v) then, for some c ∈ F ? , c−1 g fixes u and cg fixes w and so g acts on V0 . If g centralizes V0 , then g acts as scalar c on U0 and as the scalar c−1 on W0 (as u ∈ U0 and v ∈ V0 ). Hence CG (v) ∩ CG (V0 ) = CG (V ). (For the statement in parentheses take U0 = U and W0 = W .) (ii) Let g ∈ CG (v). Let (aij ) be the matrix of g on U with respect to the given basis. Then v = vg =

X i

ui g ⊗ wi g =

XX X X ( aij uj ) ⊗ wi g = uj ⊗ ( aij wi )g. i

j

j

i

Pd Thus wj g −1 = i=1 aij wi for each j = 1, · · ·, d. It follows that g acts on W0 through the inverse transpose matrix (aij )−t . This shows that W0 and V0 are F CG (v)-modules. Lifting the eigenvalues of g on U and on W0 to characteristic 0, we see that if ϕ is the Brauer character of CG (v) on U , then ϕϕ¯ is that on U ⊗F W0 . By Lemma 4.1a, V0 = U ⊗F W0 is a self-dual F CG (v)-module.

70


f is generated by w1 , · · ·, wd−1 . By (iii) Choose notation such that W f such that CG (W0 ) = CG (W ) hypothesis we find w = wd in W outside W f for the subspace W0 of W generated by W and w. Define v and ϕ as before, so that V0 = U ⊗F W0 is a self-dual F CG (v)-module affording ϕϕ¯ = |ϕ|2 . It remains to show that CG (v)/CG (V ) is faithful on V0 . Let g ∈ CG (v) act trivially on V0 . Then |ϕ(g)|2 = d2 and so |ϕ(g)| = d. Thus g acts on U by multiplication with some scalar c, whence on W0 via c−1 . By hypothesis there is g0 ∈ G acting as scalar multiplication with c−1 on W . Then gg0−1 ∈ CG (W0 ) = CG (W ) and so g = g0 on W and g ∈ CG (V ). ¤ Remark . Replacing G by G×Z for some suitable subgroup Z of Z(GL(W )) (acting trivially on U ) one can achieve that G induces all scalar transfomations on W . Every vector in V which is real for G × Z is real for G. Proposition 5.3c. Suppose V = TenG H (W ) for some (coprime) F Hmodule W with dim F W ≥ 2. If there is a (strongly) real vector in W for H, then there is also a (strongly) real vector in V for G. Proof. We prove the lemma for the case of real vectors, the proof for strongly real vectors being similar (and easier). Let {ti }ni=1 be a right transversal to H in G (with t1 = 1). For x ∈ G let i 7→ ix be the permutation (on the indices) induced by x, as in Sec. 1.2. Let w0 ∈ W be real for H, and let W0 be a subspace of W which is a self-dual F CH (w0 )module satisfying CH (w0 ) ∩ CH (W0 ) = CH (W ). Without loss of generality we assume that w0 6= 0 and, replacing W0 by W0 ⊕ F w0 if necessary, that w0 ∈ W0 . Let v0 = w0 t1 ⊗ · · · ⊗ w0 tn and V0 = W0 t1 ⊗F · · · ⊗F W0 tn . Then V0 is a subspace of V = W t1 ⊗F · · · ⊗F W tn . We assert that V0 is a self-dual F CG (v0 )-module satisfying CG (v0 ) ∩ CG (V0 ) = CG (V ). Qn Let x ∈ CG (v0 ). Then there are scalars ci ∈ F such that i=1 ci = 1 −1 and w0 ti x = ci w0 tix for each i. Hence x acts on V0 , and c−1 i ti xtix ∈ CH (w0 ). Let Htj hxi be the distinct double cosets of G modulo (H, hxi). Suppose the hxi-orbit of Htj has size nj , and let the scalar bj be the product Q P of the scalars ci belonging to this orbit. Then j nj = n, j bj = 1, −1 nj −1 tj xnj t−1 ∈ H and b t x t ∈ C (w ) for each j. By formula (1.2e) the j H 0 j j j Brauer character χ of G afforded by V takes the value χ(x) =

Y j

θ(tj xnj t−1 j )=

Y j

nj θ(b−1 j tj x tj ),

Real Vectors

71

where θ is the Brauer character of H afforded by W . Passing from H, W to CH (w0 ), W0 (character θ0 ) we get the character χ0 of CG (v0 ) afforded Q nj by V0 . Thus χ0 (x) = j θ0 (b−1 j tj x tj ). Apply Lemma 4.1a. Suppose x ∈ CG (v0 ) acts trivially on V0 . Then i 7→ ix is be the identity −1 permutation, whence x ∈ CoreG (H). Also, for each i then c−1 acts i ti xti −1 −1 as a scalar on W0 , and centralizes w0 ∈ W0 . Hence ci ti xti ∈ CH (W0 ) ∩ CH (w0 ) = CH (W ). It follows that x acts on the ith component of V = Q TenG H (W ) as the scalar ci . Using that i ci = 1 we get that x ∈ CG (V ), as desired. ¤ 5.4. Clifford Reduction Let V be a faithful, irreducible, coprime F G-module in characteristic p. Assume there is no (strongly) real vector in V for G but that (G, V ) is a minimal counterexample in the following sense: • Whenever G0 is a central extension of a subgroup of G by a p0 -group and V0 is a F0 G0 -module for which char(F0 ) = p and dim F0 V0 < dim F V , then there is a (strongly) real vector in V0 for G0 . • Theorem 5.4. Suppose (G, V ) is a minimal counterexample in the above sense (for real or strongly real vectors). Then G has a unique minimal nonabelian normal subgroup, say E, and this is either quasisimple or of extraspecial type. Moreover E is absolutely irreducible on V , and all abelian normal subgroups of G are cyclic and central. Proof. We concentrate on real vectors, the argumentation for strongly real vectors being similar. We only make use of results holding for both kinds of vectors. It is clear that G is not abelian, because otherwise there were a regular G-orbit on V by (3.4b), and d = dim F V ≥ 2 as G is faithful on V . We proceed in several steps. (1) V is an absolutely irreducible F G-module: For otherwise embed F (properly) into the field F0 = EndF G (V ), and let L Γ = Gal(F0 |F ). Then F0 ⊗F V = σ∈Γ V0σ for some absolutely irreducible F0 G-module V0 . Then V0 is faithful and dim F0 V0 < d. Thus V0 contains a real vector v0 for G, that is, ResG submodule CG (v0 ) (V0 ) has a faithful self-dual P U , say. Let Γ0 be the stabilizer of U in Γ, and let W = σ∈Γ/Γ0 U σ and P v = σ∈Γ/Γ0 uσ . V0 and V are isomorphic as G-sets (see Sec. 2.7). It

72


follows that CG (v) = CG (v0 ) and, since field and group automorphisms commute, that W is a self-dual F CG (v)-submodule of V . This contradicts our choice of V . (2) V is a primitive F G-module: Otherwise V = IndG H (U ) for some proper subgroup H of G and some F Hmodule U . Since dim F U < d there is a real vector in U for H by the choice of V . But then by part (i) of Proposition 5.3a there is a real vector in V for G, against our assumption. (3) The irreducible constituents of ResG N (V ) are absolutely irreducible for all normal subgroups N of G: Otherwise choose N maximal such that the assertion is false. Then N 6= G by (1), and there is an irreducible submodule W of ResG N (V ) such that F0 = EndF N (W ) is a proper extension field of F . Let Γ = Gal(F0 |F ), L and let F0 ⊗F W = σ∈Γ U σ for some absolutely irreducible F0 N -module U . By (2) W is G-invariant. Hence to every x ∈ G there exists a unique σ = σx ∈ Γ such that (U x)σ ∼ = U , and the assignment x 7→ σx is a homomorphism making F0 into a “G-field”. The kernel of this Galois action of G is a normal subgroup N0 containing N . Let Γ0 be the image in Γ of L σ G (so that G/N0 ∼ is an irreducible F N0 = Γ0 ). Then W0 = σ∈Γ0 U ∼ module with EndF N0 (W0 ) = F0 . Hence N0 = N by the choice of N and so G/N ∼ = Γ. It follows that F0 ⊗F V ∼ = IndG N (U ) and that ResG N (V ) = W is irreducible. We have dim F0 U < d. By the choice of V there is a real vector in U for N , and thus there is a real vector in F0 ⊗F V for G by part (i) of Proposition 5.3a. As in (1) we get a real vector in V for G. From (1), (2), (3) it follows that every abelian normal subgroup of G is cyclic and central in G (acting by scalar multiplications). Without loss of generality we may assume that G induces all scalar multiplications on V . The generalized Fitting subgroup of G is nonabelian for otherwise G were cyclic and so had a regular orbit on V . (4) ResG N (V ) is absolutely irreducible for all nonabelian normal subgroups N of G: Otherwise, in view of (3), there is a nonabelian normal subgroup N of G such that the restriction of V to N is a proper multiple eU of some

Real Vectors

73

absolutely irreducible F N -module U (e ≥ 2). This U is faithful. Let θ be the Brauer character of N afforded by U , and let G(θ) be the extended b representation group of θ. By Theorem 1.9c there is an F G(θ)-module U extending U (in the usual sense), and there is an F G(θ)-module W such b < d and b ⊗F W (viewed as an F G(θ)-module). Thus dim F U that V ∼ =U b dim F W = e < d. By the choice of V there are real vectors for G(θ) in U and in W . But then by part (i) of Proposition 5.3b there is a real vector in V for G. (5) Conclusion: Let E be a minimal nonabelian normal subgroup of G. By (4) ResG E (V ) = W is absolutely irreducible. In view of (2), (3) CG (E) = Z = Z(G) and EZ is the generalized Fitting subgroup of G. Suppose first that E is solvable. Then it is a q-group of “symplectic type” for some prime q 6= p. Either q is odd and E = Ω1 (EZ) is of exponent q, or E is a 2-group of extraspecial type. Also, E is the unique minimal nonabelian normal subgroup of G. This is clear when q is odd or q = 2 for some m. Note that and |Z(E)| = 4 (E of type 0). Suppose E ∼ = 21+2m ± G is irreducible on E/Z(E) as each proper G-invariant subgroup of E is cyclic and central in G. Thus E cannot be dihedral of order 8, and E is unique except possibly when |Z| is divisible by 4. But then E is the unique minimal nonabelian G-invariant subgroup of E ◦ Z4 . Let E be nonsolvable. Then E is the central product of the distinct G-conjugates of some quasisimple group E0 , and W is the tensor product of n = |G : NG (E0 )| distinct G-conjugates of the unique absolutely irreducible constituent W0 of ResE E0 (W ). Assume E 6= E0 (n > 1). Let θ, θ0 be the Brauer characters of N , N0 afforded by W , W0 , respectively, and let G(θ), G0 (θ0 ) be the extended representation groups. So G0 (θ0 ) is a central extension of G0 by a cyclic p0 -group. By Theorem 1.9c c0 extending W0 (in the usual sense). Since there is an F G0 (θ0 )-module W √ 1 c c0 dim F W0 = dim F W0 = (dim F W ) n = n d < d, there is a real vector in W G for G(θ0 ). Since ResN (V ) = W , we have G(θ) = G. By Theorem 1.9d there b of G by an abelian p0 -group, containing a subgroup is a finite extension G b 0 mapping onto G0 (θ), such that G b (W c V = TenG b0 0 ) G (We may arrange matters such that the Clifford correspondent of θ is trivial.) By Proposition 5.3c there is a real vector in V for G, a contradiction. Hence E = E0 is quasisimple. ¤

74


5.5. Reduced Pairs The group G is said to be reduced if all abelian normal subgroups of G are central and if G has a unique minimal nonabelian normal subgroup, E, called the core of G, which is either of extraspecial type or is quasisimple. If G is reduced with core E and V is a faithful, coprime F G-module, then (G, V ) is a reduced pair (over F ) provided E is absolutely irreducible on V . Then CG (E) = Z(G) acts as a group of scalar multiplications on V and Z(G)E is the generalized Fitting subgroup of G. We say that the pair (G, V ) is “large” if Z(G) ∼ = F ? , that is, if G induces all scalar transformations on V . The pair (G, V ) is nonreal reduced if it is reduced and if there is no real vector in V for G. The minimal counterexample (G, V ) described in Theorem 5.4 is a nonreal reduced pair. The objective of the next two chapters is to give a complete classification of such pairs. In the quasisimple case we even shall describe all reduced pairs (up to isomorphism) admitting no regular vectors. e Ve ) over F are isomorphic We say that two reduced pairs (G, V ) and (G, e if there is a group isomorphism α : G → G making Ve into an F G-module e the Brauer characters isomorphic to V . In other words, identifying G = G afforded by V and Ve are conjugate under an automorphism α of G. It is immediate that this notion of “isomorphism” for reduced pairs preserves (regular) orbits, (strongly) real vectors, and abelian vectors. 5.6. Counting Methods The approach to the classification theorems generally is in two steps: First we use counting arguments in order to reduce the discussion to “small” groups, often Atlas groups. Then we proceed by a case-by-case analysis of the remaining groups. Sometimes we use the computer, and then we refer to [Groups, Algorithms, and Programming, 2006]. This will be briefly cited as [GAP]. Let us describe the basic methods. Suppose (G, V ) is a reduced pair over F = Fr , r being a power of the prime p not dividing |G|, and let d = dim F V . Embed G into GL(V ), and let G0 = NGL(V ) (E). Then Z = Z(G0 ) = CG0 (E) ∼ = F ? is cyclic of order r − 1. Assume (G, V ) is large ¯ 0 = G0 /Z and G ¯ = G/Z, which are (replacing G by GZ, if necessary). Let G

Real Vectors

75

subgroups of Aut(E) (as Z ∩ E = Z(E)). For each v ∈ V ] , CG (v) ∩ Z = 1 and CG (v)Z/Z = CG¯ (Zv). Hence there is a 1-1 correspondence between ¯ the G-orbits on V ] and the G-orbits on the projective 1-space P1 (V ), with corresponding point stabilizers. Define the bottom β(G) of G (and similarly for each finite group) as the set of all noncentral subgroups of G of prime order. If there is v ∈ V such that CG (v) 6= 1, the stabilizer contains at least one subgroup in β(G) S (Sylow). Hence if γ∈β(G) CV (γ) 6= V , there is a regular vector in V for P G. Usually we show that γ∈β(G) |CV (γ)| < |V |, ignoring zero spaces or possible intersections. But sometimes this will be improved using common diagonalization, or arguing “projectively”, because if (5.6a)

P ¯ γ ¯ ∈β(G)

|CP1 (V ) (¯ γ )| < |P1 (V )|,

then there is a regular G-orbit on V as well. Arguing in this manner, ¯ choose a subgroup γ of G of smallest possible for any given γ¯ ∈ β(G), order mapping onto γ¯ . Then γ is a cyclic group of prime power order. If G0 = X ◦ Z is a central product over Z(E), clearly we may pick γ as a subgroup of X, and then γ either has prime order or γ ∩ Z(E) 6= 1. Let β ∗ (G) denote the set of all noncentral subgroups of G which are cyclic of order 4 or of odd prime order. Then (5.6b)

S γ∈β ∗ (G)

CV (γ) 6= V

implies that there is some v ∈ V such that CG (v) is an elementary abelian 2-group. Then v is a strongly real vector for G (and CG (v) has a regular orbit on V by Eq. (3.4b)). For a set ω of primes we denote by βω∗ (G) the set of ω-subgroups in β ∗ (G); define βω (G) similarly. If for instance S 0 γ∈βω (G) CV (γ) 6= V , there is v ∈ V such that CG (v) is a ω -group. Let g ∈ G0 . We define the fixed point ratio by (5.6c)

f (g) = f (g, V ) = dim F CV (g)/dim F V .

Of course f (g) = f (hgi) depends only on (the generators of) the group hgi. For z ∈ Z = F ? the fixed space CV (z −1 g) = Vz (g) is the eigenspace of g on V to the eigenvalue z, or it is zero. (It is for instance zero if the L order o(z) of z does not divide o(g).) We have z∈Z Vz (g) ⊆ V , with equality if and only if g is a p0 -element and F is large enough (containing

76


all eigenvalues of g). Note that g is faithful on V . There are at most d = dim F V distinct eigenspaces of g on V . Hence if f (zg) ≤ f for all S z ∈ Z then | z∈Z CV (zg)| ≤ d · rdf . The following estimate will be applied very often. Let g ∈ G0 . Suppose again that f (zg) ≤ f for all z ∈ Z, and assume that 21 ≤ f < 1. Then if zi ∈ Z are the distinct eigenvalues of g on V , (5.6d)

P i

|CV (zi−1 g)| ≤ rbdf c + rd−bdf c ≤ 2rdf .

In order to verify this, without loss of generality we may assume that g is a p0 -element and F is large enough, so g has the (distinct) eigenvalues zi Pn on V with the multiplicities di ≥ 1, 1 ≤ i ≤ n, such that i=1 di = d. Arrange these eigenvalues such that d1 ≥ d2 ≥ · · · ≥ dn . By assumption d > df ≥ d/2 and df ≥ bdf c ≥ d1 . Suppose d0j is another such decreasing Pn0 sequence of positive integers for j = 1, · · ·, n0 , satisfying j=1 d0j = d and Pn0 Pn 0 df ≥ d01 . Then i=1 rdi ≤ j=1 rdj if and only if (d1 , d2 , · · ·) ≤ (d01 , d02 , · · ·) Pn in lexicographical ordering. Consequently i=1 rdi ≤ rbdf c +rd−bdf c , which is at most 2rbdf c if f > 21 , and at most 2rdf if f = 21 . Notation. Suppose g has on V the eigenvalues zi with multiplicities di , 1 ≤ i ≤ n. Arrange these such that d1 ≥ d2 ≥ · · · ≥ dn . Then we indicate the spectral pattern by writing (d1 )

gV = [z1

, · · ·, zn(dn ) ] = [d1 , · · ·, dn ],

the latter (weak form) in the case when only the dimensions of the eigenPn spaces are of interest. Thus i=1 di ≤ d = dim F V , and we have equality if and only if g is a p0 -element of order dividing r − 1. It is obvious but important in applications that each element of G0 in the conjugacy class of g has the same spectral pattern of the first kind, and each element in the coset Zg has the same weak pattern. In particular, |CP1 (V ) (Zg)| = Pn 1 di i=1 (r − 1). r−1 Now we turn to character theory. Let χ be the Brauer character of G0 afforded by V , and assume we know χ (to some extent). Let g ∈ G0 be a 0 p0 -element of order s, say. Then V is a projective F hgi-module, ResG hgi (χ) an ordinary character and so (Sec. 1.3) (5.6e)

dim F CV (g) = hχ, 1ihgi =

1 s

Ps i=1

χ(g i ).

Real Vectors

77

Hence the fixed point ratio f (g) is just the average over the ratios χ(g i )/χ(1) for i = 1, · · ·, s. Recall from Sec. 1.5 that the χ(g i ) for the generators g i of hgi are the algebraic conjugates of χ(g). If for instance s is a prime, therefore the strong pattern gV is determined by χ(g) since the sth roots of unity 6= 1 are linear independent over Z (and their sum equals −1). We can compute the spectral pattern from the character table (in terms of F = Fr ). 5.7. Two Examples To illustrate the counting methods we discuss in some detail two reduced pairs, one of extraspecial type and one of quasisimple type. Example 5.7a. Let (G, V ) be a large reduced pair over F = Fr with core E ∼ = 51+2 + . Since E is faithful and absolutely irreducible on V , we have char(F ) = p 6= 5, d = dim F V = 5 and 5 | r − 1. Hence r ≥ 11. Embed G into GL(V ) = GL5 (r), and consider G0 = NGL(V ) (E). From Theorem 4.3c and Proposition 4.3a it follows that G0 = X ◦ Z is a central product over Z(E), where X is the standard holomorph of E and where Z = Z(G0 ) ∼ = F ? . We also know that X = E : S is a semidirect product where the complement S ∼ = Sp2 (5) is determined up to conjugacy (under E). We are going to show that there is a regular vector in V for G. By definition of a reduced pair G is irreducible on U = E/Z(E) = EZ/Z. If p = 2 or 3, then G/EZ has order 3 or is a quaternion group of order 8, respectively, and it is easy to show that there is a regular G-orbit on V . So let p > 5. Then G0 is a p0 -group, and it is enough to consider G = G0 . (Usually we are treating this “worst” case ignoring whether G0 is a p0 -group or not, counting scalar products with the 1-character “symbolically” as dimensions of fixed spaces.) Let χ be the Brauer character of X (and of G) afforded by V . This χ is one of four faithful irreducible characters of degree 5 of the holomorph X which, however, are conjugate under automorphisms of X by Theorem 4.3c, so lead to isomorphic reduced pairs. Therefore we shall sometimes speak of “the” character associated to X. By Theorem 4.5a, ResX S (χ) = ξ is “the” generic Weil character of S. In the terminology of the [Atlas, p. 2], ξ = χ2 + χ7 or χ3 + χ6 . ¯ = X/Z(E) with G/Z. Each prime order element We may identify X ¯ Z(E)x of X is represented either by a (noncentral) element of E (of order

78


5) or lies in a coset Ex for some element x ∈ S of prime order. There are (52 − 1)/4 = 6 subgroups of order 5 in U = E/Z(E), their generators having weak spectral pattern [1(5) ] as χ vanishes on noncentral elements 6 · 5(r − 1) = 30 points in P1 (V ), or 6(5r) = 30r of E. Thus at most r−1 vectors in V , are fixed by these subgroups. There are two conjugacy classes √ 5 5A0 B0 of elements in S of order 5, and if y belongs to 5A , then ξ(y) = 0 √ 1 −1 −1 (and y belongs to 5B0 with ξ(y ) = − 5). There are 2 |5A0 B0 | = 6 (conjugate) subgroups in S of order 5. By Theorem 4.4, y is good for U and |E : CE (y)| = 5. We compute that dim F CV (y) = 1 and (2)

(2)

yV = [ε1 , ε2 , 1(1) ] = [2, 2, 1] for some primitive 5th roots of unity ε1 6= ε2 which are inverse to each other (5 | r − 1 and ε1 + ε2 is a square root of 5 in F ). By Theorem 4.4 the coset Ey contains just |U : CU (y)| = 5 good cosets of Z(E) (which are conjugate under E). So there are 52 − 5 = 20 bad cosets of Z(E) in Ey. If an element y0 in Ey is bad, then χ(y0 ) = 0 and so o(y0 ) = 5 and (y0 )V = [1(5) ] (weak pattern). We therefore have X ¡ ¢ |CV (γ)| ≤ 30r + 6 5(2r2 + r) + 20(5r) . γ∈β5 (G)

¯ have order 3, and let x ∈ S be such that x Let x ¯∈X ¯ is contained in Ex. Then x belongs to the unique conjugacy class 3A0 of elements of order 3 in S. It determines the coset Ex in X, which in turn determines the coset ZEx in G. There are |S : CS (x)| = |3A0 | = 2·10 cosets of E in X conjugate to Ex under S. By Theorem 4.4 the coset Z(E)x is good for U , and the good cosets lying in Ex are conjugate under E (and so have the same weak spectral pattern). From χ(x) = −1 we deduce that |CU (x)| = χ(x)2 = 1, hence |U : CU (x)| = |U | = 52 , and that (2)

(2)

xV = [z1 , z2 , 1] = [2, 2, 1] if 3 | r − 1, where z1 6= z2 are primitive 3rd roots of unity, and xV = [1] otherwise (dim F CV (x) = 1). It follows that in Ex there are just 52 −5 = 20 bad cosets Z(E)x0 . Since χ(x0 ) = 0, the bad element x0 cannot have (prime) order 3. But each coset of Z(E) in X of order 3 can be represented by an element of order 3. Hence we may ignore the bad cosets. We conclude that X |CV (γ)| ≤ 10 · 52 (2r2 + r) γ∈β3 (G)

Real Vectors

79

if 3 | r − 1, and we may replace 2r2 + r by r otherwise. Let j ∈ S represent the unique (central) involution in S (belonging to 1A1 ). We have |CU (j)| = χ(1)2 = (−1)2 = 1 and so |U : CU (j)| = 52 . P Arguing as before we get γ∈β2 (G) |CV (γ)| ≤ 52 (r3 + r2 ). Consequently P 3 2 5 γ∈β(G) |CV (γ)| ≤ 25r + 585r + 940r. This is less than |V | = r for r ≥ 10. Hence there is a regular G-orbit. Example 5.7b. Suppose (G, V ) is a large reduced pair over F = Fr , with core E = 2.A6 , and d = dim F V = 4. Then p ≥ 7 by coprimeness, and V affords, as an F E-module, one of the characters χ8 , χ9 as a Brauer character (Atlas, p. 5). These characters are the Weil characters ξ1 , ξ1¦ of Sp2 (9) in the terminology of Theorem 4.5a. So they are rational-valued and fuse in 2.A6 .22 . But we are in an exceptional situation: The characters extend√to √ 2.A6 .21 , requiring ±3. More precisely, χ8 extends to 2+ S6 requiring 3 √ and to 2− S6 requiring −3, and for χ9 it is vice versa. Recall that 2+ S6 is that covering group of S6 in which transpositions lift to involutions (and for which the characters are given in the Atlas). In the following we assume that χ = χ9 on E, for convenience. Exchanging the conjugacy classes 3A, 3B leads from one character to the other. Similarly, exchanging the conjugacy classes 6A, 6B leads from one isoclinic variant of 2.A6 .21 to the other. So the arguments will carry over. Embed G into GL(V ), and let G0 = NGL(V ) (E). Then Z = Z(G) = CG0 (E) ∼ = F ? . In the counting argument we assume (implicitly) that F contains a square root of 3 or −3 (or both, which happens when 4 | r − 1). √ Let us consider the (worse) case that −3 ∈ F and that G = G0 = X ◦ Z where X = 2+ S6 . We ask whether there is a regular G-orbit on V , or a real vector in V for G, at least. Let x ∈ E be an element of order 3 belonging to 3A0 (mapping onto 3-cycles in S6 ). Then χ(x) = 1 and xV = [1(2) , z (2) ] for some primitive 3rd root of unity z (viewed as an element of Z). In particular W = CV (x) has dimension 2. If an element in E fixes a nonzero vector in V , then it is conjugate to x [Atlas]. Let y ∈ CE (x) be an element of order 3 in the class 3B0 (mapping onto double 3-cycles). Then χ(y) = −2 and so yV = [z (2) , z¯(2) ]. Of course W is stable under y, and we have yW = [z (1) , z¯(1) ].

80


In order to see this, we have to exclude that y acts as a scalar multiplication on W . In that case, xy and xy 2 would have nontrivial fixed points on V by common diagonalization. But xy and xy 2 do belong to the class 3B0 . We have |3A0 | = |3B0 | = 40. Since these are the unique conjugacy classes of P 40 2 2 order 3 in E and in G, we have γ∈β3 (G) |CV (γ)| ≤ 40 2 (r + 2r) + 2 (2r ) = 2 60r + 40r. On the two conjugacy classes 5A0 B0 of order 5 in E the character χ = χ9 takes the value −1 and so the weak pattern is [1, 1, 1, 1] if 5 | r − 1 and empty otherwise. From |5A0 | = |5B0 | = 72 we see that there are just 1 |5A0 5B0 | = 72/((5 − 1)/2) = 36 subgroups of order 5 in E. Hence P 2 γ∈β5 (G) |CV (γ)| ≤ 36(4r). Let g ∈ X be an element of order 6 in the class 6A0 such that g 2 = x. √ Then χ(g) = ± −3, and j = g 3 is an involution in the class 2B0 (mapping onto transpositions), which satisfies χ(j) = 0. If an element in X r E fixes a nonzero vector in V , then it is conjugate to j [Atlas]. Hence jV = [1(2) , −1(2) ] and gW = jW = [−1(2) ], because g = jx−1 = x−1 j has no fixed points on V ] . The conjugacy classes 2A and 2C in S6 lift to classes of elements of order 4 in X, and these have to be considered when 4 | r − 1. Using that χ vanishes also on 2A0 and 2C0 , we then get the contribution (|2B| + |2A| + |2C|)(2r2 ) = (15 + 45 + 15)(2r2 ) = 150r2 . There is a regular G-orbit on V provided (60r2 + 40r) + 144r + 150r2 < r4 = |V |. This is true when r ≥ 16. Hence r = 7, 11 or 13. We cannot have √ r = 11 since 3 | r − 1; we can exclude r = 11 also when assuming that 3 ∈ F using that then 4 - r − 1, replacing the summand 150r2 by 30r2 . One can also show that there is a regular G-orbit on V when r = 13. Either one argues as in the r = 7 case below (where there is no regular orbit, however), using that dim F W = dim F CV (x) = hχ, 1ihxi is independent of F , or applying the method to be developed in Proposition 7.3c below. So let r = 7 in what follows. For each nonzero vector v ∈ V we have CE (v) = hxi if v ∈ W = CV (x) and CE (v) = 1 otherwise (see above). So

Real Vectors

81

we have E-orbits on V ] of size |E : hxi| = 240 and regular orbits. Let ∆ be an E-orbit of size 240. Then ∆ ∩ W 6= ∅ and |∆ ∩ W | = |NE (W ) : hxi|. Since NE (W ) ⊇ NE (hxi) and |NE (hxi) : hxi| = 12, we can infer that NE (W ) = NE (hxi). Moreover, from |W ] | = 72 − 1 = 48 we deduce that there are exactly four E-orbits ∆i of size 240, each satisfying |∆i ∩ W | = 12 (0 ≤ i ≤ 3). From |V ] | = 74 − 1 = 2.400 we conclude that there are just two regular E-orbits Ω1 , Ω2 . Write −j for the product of j with the generator of Z(X). We know that −j centralizes W and acts as −1 on V /W . Hence each ∆i remains an X-orbit, with point stabilizer hx, −ji ∼ belongs = Z6 . The element x(−j) √ to the class 6A0 in X and is conjugate to g. Since χ(g) = ± −3 and g is diagonal on V , from Lemma 4.1c it follows that no vector in the ∆i is real for X (as 6 does not divide 7 + 1 = 8). On the other hand, Ω1 , Ω2 fuse in X, because −j does not fix any vector in Ω1 or Ω2 and X = hE, −ji. So Ω = Ω1 ∪ Ω2 is a regular orbit for X. From yW = [z, z¯] we see that there are just 6 + 6 = 12 (nonzero) eigenvectors of y in W . The group hZ(X), yi ∼ = Z6 acts semiregularly on W ] , and if ∆i ∩W contains an eigenvector of y for some i, all six eigenvectors of y to the given eigenvalue belong to ∆i ∩ U . Hence there exists ∆i such that ∆i ∩ U does not contain an eigenvector of y. Consider the action of hyzi on V ] . Clearly Ω1 , Ω2 and Ω are fixed by yz. Hence there is vj ∈ Ωj such that CEZ (vj ) = hyzi. From Lemma 4.1c it follows that no vector in Ω is real for EZ, hence not real for G = XZ. Note again that yz acts diagonally on V and that 3 does not divide 7 + 1. Also, hyzi acts on the E-orbits ∆i , and such a ∆i remains an EZ-orbit if and only if there is wi ∈ ∆i ∩ W such that CEZ (wi ) = hx, yzi (by conjugacy of y, y −1 in E). From the observation in the preceding paragraph we infer that just one orbit, say ∆0 , is preserved by hyzi, and the other ones are fused. So ∆ = ∆1 ∪ ∆2 ∪ ∆3 is an EZ-orbit with point stabilizer hxi, and ∆0 has the stabilizer hx, yzi. The vectors in ∆ are real for EZ, those in ∆0 are not. We conclude that Ω, ∆ and ∆0 are the distinct G-orbits on V ] , with point stabilizers conjugate to hyzi ∼ = Z3 , hx(−j)i ∼ = Z6 and hx(−j), yzi ∼ = Z3 × S3 (= Z3 wr S2 ), respectively. No vector in V is real for G.

Chapter 6

Reduced Pairs of Extraspecial Type In this chapter we classify the nonreal reduced pairs of extraspecial type, up to isomorphism. In terms of their cores we have just three types (Q8 ), (25− ) and (33+ ), the pairs being defined over certain prime fields (of order 3, 5, 7 or 13). 6.1. Nonreal Reduced Pairs Throughout (G, V ) is a reduced pair over a finite field F = Fr of charac1+2m teristic p where the core E ∼ is of extraspecial type (Sec. 5.5). So = q±,0 either E is an extraspecial q-group of odd exponent q 6= p and order q 1+2m , or q = 2 6= p and E is extraspecial of + or − type and order 21+2m , or E is the central product of such a 2-group with a cyclic group of order 4 (type 0). By the character theory of E we have d = dim F V = q m . Furthermore |Z(E)| is a divisor of r − 1. Without loss of generality we assume that E∼ when q = 2 and F contains the 4th roots of unity (4 | r − 1). = 21+2m 0 We now give a detailed description of certain nonreal reduced pairs, including orbit structures and point stabilizers (which will be of relevance). With one exception the pairs are large. The target will be to show that there are no further nonreal reduced pairs of extraspecial type. Type (Q8 ) : Let r = p be one of the primes 5, 7, 11 or 23. In Sec. 3.2 we have seen that the quaternion group E ∼ = Q8 embeds into GL2 (p) such that (2) G = NGL2 (p) (E) acts transitively on the nonzero vectors in V = Fp . This 0 G is a p -group. The stabilizer in G of any nonzero vector is cyclic of order 4, 3, 2 or 1, correspondingly. Complete reducibility and Lemma 4.1c imply that (G, V ) is a nonreal reduced pair when p = 5 or p = 7. For p = 5 this G is the standard holomorph of 21+2 ∼ = E ◦ Z4 (and a 0

5-complement in GL2 (5)). For p = 7 we have G = X ◦ Z6 = X × Z3 where √ X∼ with value field Q( 2) = 2− S4 is the standard holomorph of E ∼ = 21+2 − (since 2 is a square mod 7). This G has a unique subgroup G1 ∼ = Sp2 (3)×Z3 of index 2, which evidently gives rise to a further nonreal reduced pair (G1 , V ). 82

Reduced Pairs of Extraspecial Type

83

Let r = p = 13. Embed E into GL2 (13). The normalizer G = (2) NGL2 (13) (E) has two orbits on nonzero vectors in V = F13 , with cyclic stabilizers of order 4 and 3. Again (G, V ) is a nonreal reduced pair. Type (25− ) : Let GV be the largest Bucht group studied in Sec. 3.2. Here G is faithful on V and transitive on V ] , any point stabilizer H being cyclic order 8. Also, F = F3 , dim F V = 4, and G contains E ∼ = 21+4 − as a normal subgroup, which is absolutely irreducible on V (G ∼ = E.(Z5 : Z4 )). Complete reducibility and Lemma 4.1c imply that (G, V ) is a nonreal reduced pair (as 8 | 32 − 1 but 8 - 31 + 1). Let r = p = 7. Embed E ∼ = 21+4 into GL4 (7). Then G = NGL (7) (E) −

4

is isomorphic to X ◦ Z6 where X is the standard holomorph of E with value √ field Q( 2). Here G is a 7 0 -group and has two orbits on nonzero vectors (4) in V = F7 , with point stabilizers Z6 resp. SL2 (3). No vector in V is real for G (cf. Prop. 6.6b below). Similar statement for the unique subgroup G1 = Y ◦ Z6 of index 2 in G, where Y /E ∼ = Ω− 4 (2). Type (33+ ) : Let X = E : Sp2 (3) be the standard holomorph of E = 31+2 + , and let χ be one of the faithful irreducible characters of X of degree 3, which by Theorem 4.3c are conjugate under automorphisms of X. In view of the discussions in Theorem 4.5a there is, up to conjugacy in X, a unique complement S ∼ = Sp2 (3) to E in X such that ResX S (χ) = 1S + ξ2 contains the 1-character. Here ξ2 is an irreducible Weil character of degree 2, with √ Q(ξ2 ) = Q( −3). Let r = p = 7 or 13. Let V be a (coprime) F X-module affording χ as a Brauer character. Embed E and X into GL(V ) through χ, and let G = NGL(V ) (E) = X ◦ Zr−1 (Proposition 4.3a). For r = 7 there are three (2) (2) G-orbits on V ] with point stabilizers S.1, Z3 .2 = Z6 and Z3 .2 = Z3 × Z2 (where A.B indicates that A is the stabilizer taken in X). It follows that both (X, V ) and (G, V ) are nonreal reduced pairs. For r = 13 we have six G-orbits on V ] , with point stabilizers S.1 (four X-orbits fusing in G), (2) (2) Z3 .2 = Z6 (3 times), Z3 .2 = Z3 × Z2 and Z1 .4 = Z4 (one regular X-orbit). Just the pair (G, V ) is nonreal reduced in this r = 13 case. Theorem 6.1. Up to isomorphism there are just 9 nonreal reduced pairs of extraspecial type, and these are described above. All other reduced pairs (G, V ) of extraspecial type admit a strongly real vector v ∈ V for G such that CG (v) has a regular orbit on V . This will be established in the course of this chapter.

84


6.2. Fixed Point Ratios Let (G, V ) be reduced of extraspecial type, as above. We embed the core E, and G, into GL(V ) and let G0 = NGL(V ) (E). If r is odd and 4 - r − 1, the field F contains a square root either of 2 or of −2. Hence from Theorems 4.3c, 4.3d and 4.3e we deduce, in view of Proposition 4.3a, that (6.2a)

G0 = X ◦ Z

is a unique central product over Z(E) where X is that standard holomorph of E fitting into GL(V ). (This refers only to E ∼ , where we have = 21+2m ± √ √ two standard holomorphs determined by the value field Q( 2) or Q( −2).) There is an ordinary irreducible character χ of G0 which agrees on p0 elements with the Brauer character of G0 afforded by V . This is an obvious extension to the central product of “the” character associated to X, that is, of one of the faithful irreducible characters of X of degree q m (conjugate under automorphisms of X and so leading to isomorphic reduced pairs). (6.2b)

U = E/Z(E) = EZ/Z

carries in the natural way the structure of a symplectic resp. orthogonal ¯ = X/E = G0 /ZE = G ¯ 0 may be identified with Fq G0 -module. In fact, X Sp2m (q) for odd q and for q = 2 and r ≡ 1 (mod 4), and with O± 2m (2) ¯ = GZ/ZE acts irreotherwise (the sign depending on E). By definition G ducibly on U , because E is the unique minimal nonabelian normal subgroup of G and all abelian normal subgroups are cyclic and central (contained in Z(G) = Z ∩ G). The concept of “good” elements for U applies to all (noncentral) elements of G0 ; in particular (4.4) applies. Theorem 6.2c. Suppose g ∈ G0 is a noncentral p0 -element of order s where s is a prime or s = 4. (i) Suppose q | s. Then dim F CV (g) ≤ 2q m−1 when s = q is odd, and dim F CV (g) ≤ 1s (2m + (s − 1) · 2m−1 ) otherwise. If g is not good for U , then even dim F CV (g) ≤ q m−1 . (ii) Suppose q - s. Then dim F CV (g) ≤ 1s (q m + (s − 1)q m−1 ). If s is odd and s ≥ q + 1, then even dim F CV (g) ≤ 1s (q m + (s − 1)q m−2 ). Proof. Of course we use that dim F CV (g) = hχ, 1ihgi by Eq. (5.6e).


85

(i) If g is not good for U , then χ(g) = 0 by Theorem 4.4. Then χ m vanishes on each generator of hgi and so hχ, 1ihgi = qs = q m−1 when s = q is a prime. For s = 4 (and q = 2) observe that χ(g 2 ) = −q m if g 2 ∈ Z, and otherwise χ(g 2 ) is an integer of absolute value at most q m−1 by Theorem 4.4. The result follows. Suppose g is good for U . Then |χ(g)|2 = |CU (g)| by Theorem 4.4. Since g is not central in G0 , |χ(g)|2 = q n for some n ≤ 2m − 1. Consider first the case q = 2. If s = 2, then χ(g) is an integer, hence n is even and χ(g) = ±2n/2 . It follows that hχ, 1ihgi = 21 (2m ± 2n/2 ) ≤ 21 (2m + 2m−1 ). Suppose next that g has order s = 4. Then χ(g) ∈ Z[i] (i2 = −1). If n is odd, there n−1 n−1 are signs εi such that χ(g) = 2 2 (ε1 +iε2 ). Then χ(g −1 ) = 2 2 (ε1 −iε2 ). n−1 Either the integer χ(g 2 ) = 0 or = −2 2 if g 2 ∈ E, and |χ(g 2 )| ≤ 22m−1 otherwise (Theorem 4.4). It follows that hχ, 1ihgi ≤ 41 (2m + 3 · 2m−1 ), as desired. If n is even, then n2 ≤ m − 1 and χ(g 2 ) ≤ 2m−1 , and we obtain the same estimate. Suppose that s = q ≥ 3. Then χ(g) ∈ Z[εq ] (εq = e2πi/q ). Let q be the unique prime ideal in Z[εq ] above q (totally Since q = ¯q ´ ³q ramified). and χ(g)χ(g) = |χ(g)|2 = q n , and since Q

( −1 q )q

is the (unique) ¢ n2 ¡ quadratic number field contained in Q(εq ), we infer that χ(g)/ ( −1 q )q is a q-adic integer. It follows that this is even an integer in Z[εq ]. From ¢ n2 ¡ | = 1, and using that the conjugates over Q have absolute |χ(g)/ ( −1 q )q ¢ n2 ¡ = ε is a root of unity (with value 1 likewise, we obtain that χ(g)/ ( −1 q )q ε2q = 1). If n = 2a is an even integer, then χ(g) = ±εq a , and from a ≤ m−1 it follows that hχ, 1ihgi ≤ 1s (q m + (s − 1)q m−1 ), which is not greater than q¡ ¢ Pq−1 ¡ k ¢ k −1 2q m−1 . So let n = 2a + 1 be odd. Using that k=1 q εq q q = (Gauss) we obtain that r q−1 X ¡k¢ k ¡ −1 ¢ a q = ±εq ε . χ(g) = ±εq q q q a

k=1

Pq−1 a

k Adding or subtracting εq k=1 εq = 0 we see that the maximum possible multiplicity of an eigenvalue of g on V is 2q a . Hence dim F CV (g) ≤ 2q a ≤ n−1 2q 2 ≤ 2q m−1 .

(ii) In this coprime case, g is good for U and |χ(g)|2 = |CU (g)| (Theorem 4.4). Also, E0 = CE (g) is a proper extraspecial subgroup of E, or of extraspecial type (mapping onto CU (g)). Let |E0 /Z(E0 )| = q 2n , and

86


n let θ be the unique irreducible constituent of ResG E0 (χ). Then θ(1) = q (n ≤ m − 1). By Theorem 1.6c there is a sign ± and a linear character µ of hgi such that χ(y) = ±µ(y)θ(1) for all generators y of hgi. Moreover, the sign is such that q m−n = hχ, θiE0 ≡ ±1 (mod s)

when s is an odd prime. In this case hχ, 1ihgi ≤

1 m (q + (s − 1)θ(1)), s

and the result follows. For s = 4 use again that χ(g 2 ) = −q m when g 2 ∈ Z, and apply Theorem 4.4 otherwise. Assume in addition that s is odd and s ≥ q + 1. If θ(1) ≤ q m−2 , the result follows. So let θ(1) = q m−1 (n = m − 1). Then the sign must be negative, s = q + 1, and χ(y) = −µ(y)θ(1) for all generators y. Indeed s = 3 and q = 2. If µ is the trivial character of hgi, even hχ, 1ihgi = 1 m m−1 ) = 0. If µ is not trivial, hχ, 1ihgi = 31 (2m + 2m−1 ) as s (q − (s − 1)q the sum over the 3rd roots of unity is zero. ¤ Remark . The bounds given in Theorem 6.2c apply also to the eigenspaces of g on V , replacing g by zg for the elements z ∈ Z of order dividing s = o(g). For all γ ∈ β ∗ (G) (Sec. 5.5) we have the following “generic” upper bounds for the fixed point ratios (depending only on q): f (γ) ≤ 31 (1 + 2q ) when q > 3, f (γ) ≤ 32 when q = 3, and f (γ) ≤ 85 for q = 2. Observe that q+1 1 1 3 2 2 q ≤ 4 (1 + q ) ≤ 3 (1 + q ) for q > 3. If g is an involution, then f (g) ≤ 2q . 6.3. Point Stabilizers of Exponent 2 ¯ is an irreducible p0 -subgroup of G ¯ 0 = Sp2m (q) when q is odd Recall that G ± ¯ or q = 2 and 4 | r − 1, and of G0 = O2m (2) otherwise. For small r this ¯ is a proper subgroup of G ¯ 0 . So we require some usually implies that G information on the maximal subgroups of these classical groups, but only for m ≤ 7 and q ≤ 3. The reader is referred to [Aschbacher, 1984] and [Liebeck, 1985] for thorough studies of this topic (see also [Kleidman and Liebeck, 1990]). We quote from [Dickson, 1901, pp. 94, 206]: 2

|Sp2m (q)| = q m (q 2m − 1)(q 2(m−1) − 1) · · · (q 2 − 1), m m(m−1) 2(m−1) |O± (q − 1) · · · (q 2 − 1). 2m (q)| = 2(q ∓ 1)q

This yields the estimates |Sp2m (q)| < q 2m

2

+m

± and |O2m (q)| < q 2m

2

−m+1

.


87

¯ 6= G ¯ 0 . Then |G| ¯ < q 4m+4 or G ¯ is isomorphic to a Lemma 6.3a. Assume G ¯ 0 from a certain distinguished finite list, the proper irreducible subgroup of G ¯ 0 of maximal order being known. subgroups of G This refers to Theorems 4.2, 5.2, 5.4 and 5.5 in [Liebeck, 1985]. The corresponding lists of (nonparabolic) maximal irreducible subgroups can be found in Tables 3.5.C, 3.5.E and 3.5.F in [Kleidman–Liebeck, 1990]. For ¯0 ∼ example, if G = Sp2m (2) the list consists of Sp2k (2)wr S` where m = k` with ` > 1, Sp2k (2l ).Z` where m = k` with ` prime, O± 2m (2), and of S2m+2 when m is even, and O− (2) is “the” proper irreducible subgroup of maxi2m mal order (m ≥ 3). Theorem 6.3b. There exist v ∈ V such that CG (v) is an elementary abelian 2-group except possibly when q = 2 and m ≤ 5, or when q = 3 and m ≤ 3. S Proof. Suppose V = γ∈β ∗ (G) CV (γ). In view of the estimate (5.6b) it suffices to show that then necessarily q = 2 and m ≤ 5, or q = 3 and m ≤ 3. Our estimates will be rather crude, at first. Of course, it suffices to consider only those γ where the CV (γ) are different. We use that each γ ∈ β ∗ (G) has at least two generators. Let first q > 3. Then f (γ) ≤ 31 (1 + 2q ) by Theorem 6.2c. Using that there are at most dim F V = q m distinct eigenvalues on V for each g ∈ G m m m−1 )/3 m 2m ¯ ¯ = |GZ/EZ| ≤ we get 2|V | = 2rq ≤ r(q +2q q q |G|. Since |G| 2m2 +m |Sp2m (q)| < q , this yields that r≤

³ q 3m 2

¯ |G|

´

3 2(q m −q m−1 )

2). So it The function m 7→ mqm−1 suffices to consider the case m = 1 in order to see that we have r < 8 for q ≥ 11. But |Z(E)| = q is a divisor of |F ? | = r − 1, so that we indeed have q ≤ 7. For q = 7 we get r < 16 for m = 1, and r < 3 for m = 2. Since r is a prime power congruent 1 mod 7, the only possibility is m = 1, r = 8. ¯ is an irreducible subgroup of Sp2 (7) of (odd) order 21, But in this case G and we get a contradiction (namely r < 8).

For q = 5 we get r < 29 for m = 1 and r < 7 for m = 2, and so the only possibilities are m = 1 and r = 16 or r = 11. We have seen in Example 5.7a that for m = 1 there is always a vector v ∈ V such that CG (v) is an

88


elementary abelian 2-group (even with CG (v) = 1). So we are reduced to the cases where q = 2, 3. Let q = 3. Then f (γ) ≤ f = 2/3 for each γ ∈ β ∗ (G) by Theorem 6.2c. m m−1 m−1 ¯ which gives Using (5.6d) we get 2|V | = 2r3 ≤ (r2·3 + r3 )32m |G|, r3

m−1

≤

32m ³ 1 ´ ¯ 1 + 3m−1 |G|. 2 r

¯ ≤ |Sp2m (3)| < 32m2 +m we obtain that m ≤ 5, and for m = 5 we From |G| must have r = 2. But the case q = 3, r = 2 cannot happen. It remains to rule out the case m = 4 (q = 3). The above estimate yields that we must have r ≤ 5. Then r = 4 (as q = 3 | r − 1). Now ¯ is a proper irreducible subgroup of Sp8 (3) (of odd order) and so |G| ¯ ≤ G 10 8 2 |Sp4 (9).2)| = 2 · 3 · 5 · 41 by [Liebeck, 1985, Theorem 5.2]. This bound suffices to get a contradiction. Let q = 2. Then f (γ) ≤ f = 85 for each γ ∈ β ∗ (G) by Theorem 6.2c. Note that f 2m is an integer for m ≥ 3. (For the time being we only m discuss the cases where m ≥ 6.) Applying (5.6d) we get 2|V | = 2r2 ≤ m−3 m−3 ¯ and thus (r5·2 + r3·2 )22m |G| m−3

r3·2

³ ≤ 22m−1 1 +

1 r

2m−2

´

¯ |G|.

This yields at once that we must have m ≤ 8. For m = 8 we obtain that ¯0 ∼ ¯ r = 3 (since r is odd). In this case G = O± 16 (2). Knowing that G is a ± 30 -group and that 37 is a divisor of |O16 (2)| this case is ruled out. For m = 7 we obtain that r ≤ 7. Again the case r = 7 is ruled out using the upper bound for the order of O± 14 (2) in place of that for ¯ ≤ |O− (2)|, by the comment after Sp14 (2). For r = 5 we also have |G| 14 (6.3a), and this suffices to get a contradiction. For r = 3 we use that the order of a proper irreducible (30 -) subgroup of O± 14 (2) is bounded above by 21 9 |GU7 (2)| = 2 · 3 · 5 · 7 · 11 · 43 [Liebeck, 1985, Theorems 5.4, 5.5]. We obtain a contradiction. Let m = 6. The usual estimate gives that r24 ≤ 289 , hence r ≤ 13. ¯ ≤ Since |Sp12 (2)| is divisible by each (odd) prime p ≤ 13, we have |G| − |O12 (2)| by Lemma 6.3a, and this leads to r ≤ 5. Consider r = 5. By Lemma 6.3a we are led to a study of the irreducible 50 -subgroups of O± 12 (2) and of S14 (viewing U as shortened S14 -permutation module over F2 ). In ¯ ≤ |GU6 (2)| = 215 · 38 · 5 · 7 · 11 by the theorems of the first case we get |G|


89

Liebeck quoted above, and this bound is good enough. There is no problem ¯ ≤ |S14 |. with the bound |G| So let r = 3. We get the desired contradiction (to the above inequality) if we can show that that the order of an irreducible 30 -subgroup of 27 O± 12 (2) is bounded above by 2 . Observe that 4m + 4 = 28. We argue by inspection of the list of maximal subgroups of the orthogonal groups. The maximal subgroups of O+ 12 (2) which are irreducible are either isomorphic to + + O6 (2)wr S2 , O4 (2)wr S3 or to almost simple groups for which the 30 -part of the order is less than 227 , like for GU6 (2) or O+ 6 (4).2 (cf. Table 3.5.E in [Kleidman–Liebeck, 1985]). The maximal subgroups of O− 12 (2) which − are irreducible are either isomorphic to O− (2)wr S or to O (4).2, or are 3 4 6 0 certain almost simple groups for which the 3 -part of the order is less than 227 , like for S13 (cf. Table 3.5.F in [Kleidman–Liebeck, 1990]). From this last table one infers that the 30 -subgroup of O− 6 (4).2 have the desired small order. Hence the result. ¤ Comments 6.3c. Theorem 6.3b reduces the discussion to some few cases. One can show that then there are point stabilizers of exponent 2 (or 1) when r is large enough: P Suppose q = 3 and m = 3. The crude bound γ∈β ∗ (G) |CV (γ)| ≤ 2 m 1 m 1 ¯ used in the above proof is sufficient to show, 33 + r 3 3 ) · 32m · |G| 2 (r ¯ ≤ |Sp6 (3)| = 210 · 39 · 5 · 7 · 13, that there is taking the upper bound |G| v ∈ V such that CG (v) is an elementary abelian 2-group provided r > 19. For m = 2 the corresponding holds provided r > 157, and for m = 1 whenever r > 211, but it is easy to improve these estimates on the basis of Theorems 4.4, 4.5a. See Sec. 6.5 below. Let q = 2. If m = 5, the analogous estimate, taking the upper bound ¯ |G| ≤ |Sp10 (2)| = 225 ·36 ·52 ·7·11·17·31, yields that there is v ∈ V such that CG (v) is an elementary abelian 2-group provided r > 37. In the orthogonal ¯ ≤ |O− (2)| = 221 · 36 · 52 · 7 · 11 · 17, this case, taking the upper bound |G| 10 holds for r > 23. For m ≤ 4 we argue a bit more carefully: m = 4: In Sp8 (2) there are just a = 346.832.896 elements of order at most 4 [Atlas, pp. 124, 125]. By Theorem 6.2c, |CV (γ)| ≤ r10 , r8 , r9 for γ ∈ β3 (G), β4∗ (G) or βs (G) for a prime s ≥ 5, respectively. If there is no vector v ∈ V such that CG (v) is an elementary abelian 2-group, then in view of the estimate (5.6d) 2 · r16 ≤ 28 a(r10 + r6 ) + 28 · (|Sp8 (2)| − a)(r9 + r7 ).

90


Using that |Sp8 (2)| = 216 · 37 · 52 · 7 · 17 this forces that r ≤ 71. m = 3: In Sp6 (2) there are just a = 80.707 elements of order at most 4 [Atlas, p. 47]. By Theorem 6.2c, |CV (γ)| ≤ r4 , r5 , r4 for γ ∈ β3 (G), β4∗ (G) or βs (G) for a prime s ≥ 5, respectively. From 2 · r8 ≤ 26 a(r5 + r3 ) + 26 (|Sp6 (2)| − a)(2r4 ) we get r ≤ 147. Hence there is v ∈ V such that CG (v) is an elementary abelian 2-group provided r > 139. m = 2: In Sp (2) ∼ = S6 there are just b = 480 elements of 2-power order 4

or of prime order. Hence at most 480 cosets of G0 mod Z contain noncentral elements of order 4 or of odd prime order. From 2 · r4 ≤ 24 · 480(r2 + r2 ) we get r ≤ 83. m = 1: From 2 · r2 ≤ |G0 /Z|(2r) = 48r we get r ≤ 23. The corresponding bounds for the orthogonal cases are somewhat better. We emphasize that a more careful computation yields much better estimates which, however, do not suffice for our purposes. This makes the following case-by-case analysis necessary. We attempt to avoid computer calculations, and this is possible in the present situation. We shall use both (elementary) counting methods and theoretical arguments.

6.4. Characteristic 2 Theorem 6.4. Let p = 2. Then there is a strongly real vector v ∈ V for G such that CG (v) has a regular orbit on V . Proof. Here G has odd order (by coprimeness) and so is solvable by the Feit–Thompson theorem. By virtue of Theorem 6.3b we can assume that q = 3 and that m ≤ 3 (and dim F V = 3m ). Of course, F is a field of order r where r is a 2-power and r − 1 is divisible by 3. By Theorem 4.3c there is a subgroup S ∼ = Sp2m (3) which is a complement to E in X. Let H be the subgroup of S mapping onto ¯ = GZ/EZ. We know that H is faithful and irreducible on U = EZ/Z, G whence O3 (H) = 1. Also, H has odd order. For m = 1 this forces that H = 1. Hence it remains to examine the cases m = 2, 3. By Theorem 4.5a 0 ResX S (χ) = ξ is a Weil character of S and ξ = ξ1 + ξ2 = 2ξ1 + 1S on 2 elements of S, where the ξi are irreducible characters of S. Thus there is v in V ] such that CS (v) ⊇ H (by coprimeness). We have NE (F v) 6= E as E is


91

irreducible on V . Moreover, NE (F v) is H-invariant and so NE (F v) = Z(E) as H is irreducible on U . It follows that NHE (F v) = H × Z(E). Hence CGZ (v) = CHE (v) = H by Lemma 5.1b, and v is strongly real for G provided ξ takes only real values on H. Recall from Theorem 4.5a that ξ takes rational values on the 30 -elements of H. Let m = 2. Then ξ = χ2 +χ21 (resp. ξ¯ = χ3 +χ22 ) in the Atlas notation [Atlas, p. 26]. Checking the list of (irreducible) maximal subgroups of Sp4 (3) we see that H must be in a group of the form 2.(24 : A5 ) or of the form 2.S6 . In both cases necessarily |H| = 5 and ξ(x) = −1 for each generator x of H (belonging to the conjugacy class 5A). Let m = 3. The irreducible odd order group H can only be contained in (almost maximal) subgroups of Sp6 (3) of the following types: 2.A5 , Sp2 (13), SL3 (3), U3 (3).2, Sp2 (3)wr S3 and Sp2 (27) : 3 [Atlas, p. 113]. Considering the maximal (and second maximal) subgroups of these groups we get that either H is cyclic of order 7 or is a Frobenius group of order 21. Note that H cannot be of order 13 or of type 13 : 3, because 7 is the unique Zsigmondy prime divisor of 36 − 1 which divides 33 + 1 = 28, the order of a Singer cycle in Sp6 (3). This implies that H is contained in groups of types Sp2 (13) or PGL2 (7) (embedded in U3 (3).2), which are real groups, or in a group of type Sp2 (27) : 3 (but not in Sp2 (27) unless |H| = 7). Also, H is uniquely determined in Sp6 (3) up to conjugacy. By the character table of Sp6 (3) the elements h ∈ H of order 7 have ξ(h) = −1 and dim F CV (h) = 3. We may assume without loss that H is Frobenius of type 7 : 3. We have to show that ξ takes only real values on the elements of order 3, and that there is a regular H-orbit on V . Inspection of the character tables of Sp2 (13), U3 (3).2 ∼ = G2 (2) and Sp2 (27) : 3 yields that we must have ξ(y) = 0, 3 or −9 for each element y ∈ H of order 3, as desired. Since the fixed spaces of h or y on V have dimension at most 11, and since 21 · r11 < r27 for r ≥ 4, there is a regular H-orbit on V . ¤ Remark . Theorem 6.4 solves the k(GV ) problem in characteristic 2, in view of our previous results. The Robinson–Thompson Theorem 5.2b tells us that it suffices to find a (strongly) real vector in V for G. Arguing by induction on dim F V by Theorem 5.4 we are led to the situation that (G, V ) is a reduced pair, and then Theorem 6.4 applies. A different approach (for groups G of odd order) has been given in [Gluck, 1984].

92


6.5. Extraspecial 3-Groups Theorem 6.5. Suppose p and q are odd. Then there exists a strongly real vector v ∈ V such that CG (v) has a regular orbit on V , except when (G, V ) is isomorphic to one of the three nonreal reduced pairs of type (33+ ) for r = p = 7, 13. Proof. By Theorem 6.3b we may assume that q = 3 and m ≤ 3. Thus p ≥ 5 and 3 | r − 1, whence r ≥ 7. By Theorem 4.3c, X = E : S is a semidirect product of E ∼ with S ∼ = 31+2m = Sp2m (3), the complements + being conjugate unless m = 1 (in which case S · Z(E)/Z(E) is determined up to conjugacy in X/Z(E)). Also, each element of S · Z(E) is good for U = E/Z(E). Recall also that ResX S (χ) = ξ is a Weil character of S (Theorem 4.5a). Let j be the central involution of S, which inverts the elements of U . Let m = 1. Then G0 is a p0 -group. Without loss of generality let G = G0 . The character table of S ∼ = Sp2 (3) can be found in [I, p. 288]. Using the notation used there ξ = χ2 + χ7 (resp. ξ¯ = χ3 + χ6 ). We apply the counting techniques described in Sec. 5.6. There is one conjugacy class of elements x in S of order 4 with |S : CS (x)| = 2 · 3. We have ξ(x) = 1 and hence xV = [1, i, −i] if 4 | r − 1 and xV = [1] otherwise, and |U : CU (x)| = 32 . There are two conjugacy classes of elements of order 3 in S, √ −1 represented by y and y where ξ(y) = i 3 say. So yV = [z (2) , 1] for some primitive 3rd root of unity z, and |E : CE (y)| = 3 (and |S : CS (y)| = 4). By Theorem 4.4, E permutes the good cosets of Z(E) in Ey transitively by conjugation, and so there are just 32 − 3 = 6 cosets of Z(E) in Ey which are not good for U . Also, χ vanishes on the elements of order 3 in these bad cosets (so that each eigenvalue has multiplicity 1). Similar statement for the noncentral elements of E (of order 3). Since there are four subgroups of order 3 in U = E/Z(E), we get X |CV (γ)| ≤ 3 · 32 (3r) + 4 · 3(r2 + r) + (4 · 6 + 4)(3r). γ∈β ∗ (G)

This is less than |V | = r3 provided r > 20. For r = 19 we may replace the P first summand on the right by γ∈β ∗ (G) |CV (γ)| ≤ 3 · 32 r (as 4 - 19 − 1). 4 This gives the result for r = 19. The remaining cases are r = 13 and r = 7, which lead to the nonreal reduced pairs of type (33+ ). Let m = 2. Now S ∼ = Sp4 (3) has order 27 · 34 · 5. For p 6= 5 we ¯ is an irreducible (solvable) may assume that G = G0 , and otherwise G


93

50 -subgroup of Sp4 (3). We have ξ = χ2 + χ21 (resp. ξ¯ = χ3 + χ22 ) in the Atlas notation [Atlas, p. 27]. P The usual counting method yields that γ∈β ∗ (G) |CV (γ)| < |V | = r9 for r ≥ 10, but not for r = 7. An even more precise estimate gives this inequality also for r = 7, but this requires the computer [GAP]. So we only sketch the estimate, and then proceed by giving the (theoretical) argument proposed in [Gluck–Magaard, 2002a] for m = 2 and m = 3. There three conjugacy classes 2B0 and 4A0 , 4A1 of elements of order 4 in S ∼ = Sp4 (3) lead, on the basis of (4.4), to the estimate X |CV (γ)| ≤ 34 · 135(r3 + 3r2 ) + 32 · 270(3r3 ) + 34 · 270(2r3 + r2 + r). γ∈β4∗ (G)

P 4 2 Similarly we obtain γ∈β5 (G) |CV (γ)| ≤ 1.296 · 3 (4r + r). There are four conjugacy classes 3A0 B0 and 3C0 , 3D0 of elements of order 3 in S, and there are 40 subgroups of order 3 in U . This yields the upper bound P 6 3 2 4 2 5 2 γ∈β3 (G) |CV (γ)| ≤ 3 · 40(r + r ) + 3 · 120(2r + r) + 3 · 240(r + 2r ) + 4 4 2 4 2 3 (40 + 40(3 − 3) + 120(3 − 3 ) + 240(3 − 3 ))(3r ). Now sum up. In what follows let either m = 2 or m = 3. Let P be the stabilizer in S of a maximal totally isotropic subspace W of U . Thus P = R : L is a maximal parabolic subgroup of S where W is the standard module for L ∼ = 2 Sym (W ) is an irreducible LGLm (3) and where the unipotent radical R ∼ = module and L0 -module (Appendix B and Appendix A7). Also U = W ⊕W ∗ where the dual L-module W ∗ is not isomorphic to W by Lemma 4.1c. For m = 3 we know that W and W ∗ are not even isomorphic as modules for L0 ∼ = PSL3 (3) (L = L0 ×hji). For m = 2 the modules W = he1 , e2 i = SL3 (3) ∼ and W ∗ = he∗1 , e∗2 i are isomorphic L0 -modules, but the unique further (Lconjugate) irreducible L0 -submodules he1 + e∗2 , e2 − e∗1 i and he2 + e∗1 , e∗2 − e1 i of U are nondegenerate. In terms of this hyperbolic basis, let τ ∈ S be the linear (symplectic) map on U sending ei to e∗i and e∗i to −ei (as in Theorem 4.5b). Then τ 2 = j and, at any rate, NS (L0 ) = NS (L) = hL, τ i. Of course τ interchanges W and W ∗ . Each element of L is conjugate to its inverse within NS (L). Hence χ = ξ is real-valued on L. In view of (A2) H1 (L, W ) = 0 = H1 (L, W ∗ ) (as j ∈ L). Identifying W = Hom(W, Z(E)), and W = (W ∗ )∗ , therefore by (A7) there exist Linvariant elementary abelian subgroups A, A∗ of E mapping isomorphically onto W, W ∗ , respectively. Both CV (A) and CV (A∗ ) are 1-dimensional, and distinct. Let Ve = CV (A) ⊕ CV (A∗ ). Since both direct summands of Ve are ∗

94


L-invariant, L0 acts trivially on Ve . Also, τ interchanges A = [A × Z(E), L] and A∗ = [A∗ × Z(E), L], hence also CV (A) and CV (A∗ ). There is v ∈ Ve outside CV (A) and CV (A∗ ) which is not an eigenvector of τ (as r ≥ 7). Let H = CG0 (v). We assert that H ∩ ZE = 1. Otherwise, since CZ (v) = 1, H ∩ZE is a nontrivial L0 -invariant elementary abelian subgroup of ZE whose image in U = ZE/Z is nonzero and totally singular and so must be W or W ∗ . Since H contains [H ∩ ZE, L], which is A or A∗ , it follows that A or A∗ centralize v. This contradicts the choice of v. Hence the assertion. We know that NG0 (L0 ) = NG (L) = hL, τ i × Z. Assume H ⊆ NG0 (L0 ). Using that H ∩ ZE = 1 and that hL, τ i/L0 is elementary abelian (of order 4) then even H ⊆ L × h−1i. In this case v is a strongly real vector for G. Also, CG (v) = H ∩G has a regular orbit on V . This is immediate for m = 2 (as r ≥ 7). For m = 3 and p 6= 13 the group L = GL3 (3) is a p0 -group (p ≥ 5) and so has a regular orbit on each faithful irreducible L-submodule of V (which exists) by Theorem 7.2a below (see also the remark after this theorem). When p = 13 use a simple counting argument, noting that every involution in L0 has r15 fixed points on V , and that every element of order 3 at most r12 (character table), and that dim F CV (j) = 14. So the proof is accomplished if we can show that H ⊆ NG0 (L0 ) or, at least, that v can be chosen in V0 such that this holds for its stabilizer. At any rate we have H 0 ⊆ S. To see this use that τ 2 = j. We have j ∈ L0 , even j ∈ L00 ∼ = Q8 when m = 2, whereas j 6∈ L0 for m = 3 (in which case L = L0 × hji). Since τ interchanges CV (A) and CV (A∗ ), we get that j acts on V0 as 1 or −1. Hence j normalizes F v and so [H, j] ⊆ H. But j maps onto the central involution in G0 /ZE ∼ = Sp2m (3) and so [H, j] ⊆ H ∩ ZE = 1. It follows that indeed H 0 ⊆ CG0 (j)0 = (SZ)0 = S 0 = S. Recall that L0 ⊆ H (but τ 6∈ H). Since H0 = H 0 L0 centralizes v and ξ does not contain the 1-character, H0 is a proper subgroup of S. Suppose that H0 normalizes W . Then Λ = Hom(W, Z(E)) is an irreducible H0 module which agrees with W ∗ = W τ on L0 . We have H1 (L0 , Λ) = 0, either by (A2) since j ∈ L0 (m = 2), or by (A9), (A7) since Λ ∼ = W τ (m = 3). We claim that H0 normalizes A. For m = 2 we know that j ∈ H0 and so ExtH0 (W, Z(E)) = 0 by (A2) and (A7). For m = 3 we have either H0 ⊆ L or H0 is a subgroup of P containing the (irreducible) unipotent radical R = O3 (P ). In the latter case Λ and R are nonisomorphic irreducible H0 modules. Application of the exact inflation–restriction sequence (Appendix


95

A2) yields that H1 (H0 /R, Λ) ∼ = H1 (H0 , Λ). In fact, the sequence 0 → H1 (H0 /R, Λ) → H1 (H0 , Λ) → HomH0 (R, Λ) = 0 is exact. Now H0 /R ∼ = L or L0 and Λ ∼ = W ∗ as L-module. We conclude 1 that ExtH0 (W, Z(E)) = H (H0 , Λ) = 0, as before. It follows that there is an H0 -invariant subgroup of E mapping isomorphically onto W , and this must be A = [A × Z(E), L0 ], as claimed. Hence H0 normalizes CV (A) and H00 centralizes V0 = CV (A) ⊕ F v, whence CV (A∗ ). Clearly NE (CV (A∗ )) ⊇ A∗ Z(E), and we must have equality since |E : NE (CV (A∗ ))| ≥ dim F V = |E : A∗ Z(E)| and E is irreducible on V . Therefore H00 normalizes W ∗ likewise, which implies that H00 ⊆ L. We conclude that H0 ⊆ NS (L) = NS (L0 ) and H ⊆ NG0 (L0 ), which gives the result. A similar argument works when H 0 normalizes W ∗ . Let m = 3. Then NS (L) = Lhτ i is a maximal subgroup of S ∼ = Sp (3) 6

[Atlas, p. 113]. From τ 6∈ H we get that H 0 ⊇ L0 is contained in a maximal parabolic subgroup stabilizing either W or W ∗ . We are done in this case. Thus it remains to consider the case m = 2 (and r = 7). We inspect the overgroups of L0 in S which do not normalize L0 [Atlas, p. 26]. Let y be an element in L0 of order 3. Since ξ(y) is real and y centralizes V0 , y belongs to the class 3D0 of S (see above). It follows that L0 /hji cannot be in a maximal subgroup of type 24 : A5 of S/hji ∼ = PSp4 (3), because each element of order 3 in A5 centralizes a V4 -subgroup in the shortened permutation module 24 over F2 , whereas |CS (y)| = 2 · 54. Also, L0 cannot map into a maximal subgroup of PSp4 (3) of type 31+2 : Sp2 (3) as j ∈ L0 . So + 0 L must be contained in maximal subgroups X ∼ = 2.S6 = Sp2 (3)wr Z2 or Y ∼ of S. X permutes the two nondegenerate irreducible L0 -submodules of U , so is the unique overgroup of L0 of this kind. Further, L0 must be diagonally embedded in the base group B = L1 × L2 of X (L0 ∼ = Li ∼ = Sp2 (3)). We claim that H0 ⊇ O2 (X) ∼ = Q8 × Q8 . Otherwise H0 ∩ O2 (Li ) ⊆ Z(Li ) for each i, because L0 acts irreducibly on O2 (Li )/Z(Li ). A Sylow 3-subgroup of H0 acts on O2 (H0 )∩B ⊆ O2 (L0 )Z(X) ∼ = Q8 ×Z2 , inducing at most an automorphism group of order 3. We conclude that |H0 |3 = 3, because otherwise there is an element of order 3 in H0 r L0 centralizing L0 = hy, O2 (L0 )i. We infer that L0 has index 2 or 1 in H0 ∩ B, which is impossible by assumption. Now ResSB (ξ) = ξ1 ⊗ ξ2

96


where ξi is an (appropriate) Weil characters of Li (corresponding to a central decomposition E ∼ ◦ 31+2 = 31+2 + + ; cf. Lemma 7.7b for more details). Since each ξi is the sum of two irreducible Weil characters of degree 1 and 2, which remain irreducible on O2 (Li ), we obtain that O2 (B) = O2 (X) fixes exactly one point in P1 (V ), hence at most one point in P1 (Ve ). Suppose L0 ⊆ Y . Then Y is the unique 2.S6 subgroup of S containing L0 . This may be seen by using that PSp4 (3) ∼ = Ω5 (3) and noting that − ∼ Y /hji = O4 (3) is the stabilizer of a minus point in the orthogonal space [Atlas]. Also, since L0 embeds into X, its central quotient group centralizes a 1-dimensional plus type subspace [Atlas]. If L0 were contained in two different 2.S6 subgroups of S, the fixed space of L0 /hji ∼ = A4 on the orthogonal space is (at least) 3-dimensional and so A4 has a faithful irreducible module over F3 of dimension 5 − 3 = 2, which is not true. Now any two A4 subgroups of A6 are conjugate under Aut(A6 ), and each A4 subgroup is contained in exactly two A5 subgroups of A6 [Atlas, p. 4]. We conclude that L0 is contained in exactly two 2.S5 subgroups of Y , each of which can fix at most one point in P1 (Ve ). Each solvable subgroup of Y above L0 normalizes L0 . Consequently, if we remove from P1 (Ve ), besides CV (A), CV (A∗ ) and the eigenspaces of τ , eventually further 3 = 1 + 2 points (|P1 (Ve )| = r + 1 ≥ 8), there remains a 1-dimensional subspace F v for which CG0 (v) ⊆ L×h−1i. This completes the proof. ¤ 6.6. Extraspecial 2-Groups of Small Order Proposition 6.6a. Let E ∼ = 21+2 ±,0 . There exists v ∈ V such that CG (v) is an elementary abelian 2-group unless (G, V ) is isomorphic to one of the four nonreal reduced pairs of type (Q8 ) for p = r = 5, 7, 13. ∼ 21+2 , because then X either is dihedral Proof. We cannot have E = + (r ≡ 7 (mod 8)) or semidihedral (r ≡ 3 (mod 8)) of order 16 and cannot be irreducible on U = E/Z(E). So E ∼ = Q8 ◦ Z4 (in case 4 | r − 1). = Q8 or E ∼ ¯ is an irreducible In both cases |G0 | = 24(r − 1) and G0 /Z ∼ = S4 . Since G 0 ¯0 ∼ p -subgroup of G = S3 , we have p 6= 3, and we may take G = G0 . There are at most six noncentral cyclic subgroups γ in G of order 4 (and at most three if 4 - r − 1), each having a fixed space on V of dimension at most 1, because dim F CV (γ) ≤ 41 (2 + 3 · 1) ≤ 45 by Theorem 6.2c. There are at most eight noncentral subgroups of order 3 in G (and at most four


97

such subgroups if 3 - r − 1), each having a fixed point space of dimension at most 1 by Theorem 6.2c (with s = 3, q = 2). Hence X |CV (γ)| ≤ 14r < r2 = |V | γ∈β ∗ (G)

provided r > 14. For r = 11 the sum on the left is at most 7r. (For r = 11 we know from Sec. 3.2 that G = G0 ∼ = GL2 (3) × Z5 is transitive on V ] and that |CG (v)| = 2 for each non-zero vector v.) For r = 5, 7, 13 we get the nonreal reduced pairs of type (Q8 ) described in Sec. 6.1. ¤ Proposition 6.6b. Let E ∼ = 21+4 ±,0 . There exists v ∈ V such that CG (v) has a regular orbit on V , and this vector v is strongly real for G unless (G, V ) is isomorphic to one of the three nonreal reduced pairs of type (25− ). Proof. We treat the three types for E separately. 0 ¯ ¯ ∼ (i) Let E ∼ = 21+4 + . Then G is an irreducible p -subgroup of G0 = + 2 ∼ ∼ O (2) = S3 wr S2 = 3 : D8 . Therefore we have p 6= 3, and we may assume 4

that G = G0 . Further r − 1 is not divisible by 4 and so r ≥ 7. For r = 7 by [GAP] computation (or directly) one finds v ∈ V such that CG (v) ∼ = D12 , which is a real group and has a regular orbit on V . So let r > 7 (hence r ≥ 11). ¯ and let Let S be a Sylow 3-subgroup of X (mapping onto O3 (X)) Y = NX (S). By the Frattini argument Y E = X. Hence Y is irreducible on U = E/Z(E) and so Y ∩ E = Z(E). There are two Y -orbits on S ] of sizes 4. Let g1 , g2 in S represent these orbits, and choose notation such that |CU (g1 )| = 1 and |CU (g2 )| = 22 . In fact, g1 permutes the 6 nonsingular vectors in U in two orbits of size 3 and g2 fixes 3 of them. (Consider a central decomposition E = Q8 ◦ Q8 .) From Theorem 4.4 it follows that χ(g1 ) = ±1 and χ(g2 ) = ±2. Computation of dim F CV (gi ) = 31 (4 + 2χ(gi )) shows that χ(g1 ) = 1 and χ(g2 ) = −2. We have (g1 )V = [1(2) , z1 , z2 ] and (2) (2) (g2 )V = [z1 , z2 ], provided 3 | r − 1, z1 6= z2 being the primitive 3rd roots of unity. For each noncentral involution x in E we have xV = [1(2) , −1(2) ]. There are 9 = 15 − 6 singular points in U , hence 2 · 9 noncentral involutions in E. We have X 18r2 + |CV (g)| ≤ 18r2 + 2 · 24 (r2 + 2r) + 2 · 32 (2r2 ) < r4 = |V | γ∈β3 (G)

since r ≥ 11. Hence there is a vector v ∈ V such that H = CG (v) is a 2group satisfying H ∩ ZE = 1. It follows that H is isomorphic to a subgroup

98


of a dihedral group D8 . Hence χ takes only rational values on H, because exp (H) | 4 and 4 - r − 1. It is obvious that H has a regular orbit on V . (ii) Let E ∼ = 21+4 − . By (A9) there is a subgroup T of X such that T · Z(E) = X and T ∩ E = Z(E), and this is determined up to conjugacy. ¯0 ∼ (Here U = E/Z(E) is the projective Steinberg module for X = Ω− 4 (2).) 0 ∼ Assume T = A5 . Then inspection of the character table, on the basis of Theorem 4.4, shows that ResX T 0 (χ) = χ1 + χ2 or χ1 + χ3 or χ4 in the Atlas notation [Atlas, p. 2]. However, in these cases the character χ takes the value 1 on the conjugacy class 3A of elements of order 3 in A5 and so these elements have trivial fixed space on U by Theorem 4.4. But the elements of order 3 in O− 4 (2) stabilize singular or nonsingular points in U . Consequently T ∼ = 2.S5 and ResX T 0 (χ) = χ6 + χ7 , where χ6 , χ7 are algebraically conjugate but fuse to a rational-valued character of T . (The characters χ8 of degree 4 are ruled out by their values √ χ8 (6A0 ) = ± −3.) Let y represent the unique conjugacy class of elements in T of order 3. Then |T : CT (y)| = 2 · 10 and χ(y) = −2. Hence (2) (2) dim F CV (y) = 31 (4 − 2 − 2) = 0 and, when 3 | r − 1, then yV = [z1 , z2 ] where z1 6= z2 are the primitive 3rd roots of unity. In view of Theorem 4.4, P y is good for U and |E : CE (y)| = 22 . It follows that γ∈β3 (G) |CV (γ)| ≤ S 22 · 10(2r2 ) = 80r2 if 3 | r − 1, and γ∈β3 (G) CV (γ) = 0 otherwise. There are two conjugacy classes 5A0 B0 in T of elements of order 5, represented by g, g −1 say. Then |T : CT (g)| = 6 and χ(g) = −1. It follows that |E : CE (g)| = 24 , by (4.4), and that CV (g) = 0. Hence each nonidentity eigenvalue of g on V occurs with multiplicity 1 (if 5 | r − 1). Thus P 4 γ∈β5 (G) |CV (γ)| ≤ 2 · 6(4r) = 384r. There are five singular points in U , hence 5 · 2 noncentral involutions in ZE, each having a 2-dimensional fixed b space on V . Letting β(G) be the union of β3 (G), β5 (G) and the subgroups generated by these involutions, we therefore have the estimate X |CV (γ)| ≤ 90r2 + 384r. b(G) γ∈β This is less than |V | = r4 if r > 11. For r = 11 the right hand side may S be replaced by 10r2 + 384r. Hence there is v ∈ V outside γ∈βb(G) CV (γ) when r ≥ 11. Then H = CG0 (v) is a 2-group and H ∩ ZE = 1. So H is isomorphic to a Sylow 2-subgroup of S5 , which is dihedral of order 8. As in (i) v is strongly real for G, and CG (v) has a regular orbit on V .


99

It remains to examine the cases r = 3 and r = 7. In the former case ¯ maps into O− (4) ∼ ¯ 6= 1 as it G is a 30 -group and so G = 5 : 4, but O5 (G) 2 is irreducible. If GV is not the largest Bucht group, CG (v) has order 1, 2 or 4 for each v ∈ V ] . Then v is strongly real for G by Lemma 4.1c. Let r = 7. There is a subgroup S ∼ = SL2 (3) of T , containing the element y say. Then χ splits up on S∆Z3 Z3 the 1-character. If G ⊇ X 0 × Z3 , we obtain a vector v ∈ V ] such that CG (v) ∼ = SL2 (3), and v is not real for G. In this case there is only one further orbit on V ] , with point stabilizer Z3 or Z6 (depending on whether G = X 0 × Z3 or X × Z3 ), and Lemma 4.1c applies. Exclude now these two nonreal reduced pairs of type (25− ) just described. ¯ must map again into O− (4). The above estimate Then either G ⊆ X, or G 2 applies ignoring the elements of order 3 and 5. ∼ 21+4 (so that 4 | r − 1). By (A9) and the character (iii) Let E = 0 ¯0 ∼ table of G = Sp4 (2) ∼ = S6 there is a subgroup T ∼ = 2.A6 of X such that Z(T ) = T ∩ Z, and ResX T (χ) = χ8 or χ9 in the Atlas notation [Atlas, p. 5]. ¯ is an irreducible 30 -subgroup of Consider first the case p = 3. Then G ¯ G0 . Inspection of the maximal and second maximal subgroups of S6 [Atlas, ∼ 5 : 4. ¯ is isomorphic to a subgroup of O− (4).2 = pp. 4, 2] shows that G 2 − ¯∼ Without loss of generality we may assume that G O (4).2, in which case = 2 − ¯ the elements of order 2 (and 5) in G belong to O2 (4) ⊆ A6 . Let x ∈ T be ¯ Then χ(x) = −1 and an element of order 5 whose coset generates O5 (G). C (x) = 0 [Atlas] and |E : CE (x)| = 24 by Theorem 4.4. It follows that PV 4 e γ∈β5 (G) |CV (γ)| ≤ 2 (4r) (assuming the worse case 5 | r − 1). Let β4 (G) be the set of cyclic subgroups of G of order 4 or 2 mapping onto the five ¯ Let y¯ be an involution in G. ¯ Then U = E/Z(E) subgroups of order 2 in G. is the regular F4 h¯ y i-module. There is y0 ∈ T mapping onto y¯, and this y0 has order 4 and satisfies y02 ∈ Z(E) and χ(y0 ) = 0 [Atlas]. Thus y0 is bad for U , CV (y0 ) = 0 and (y0 )V = [i(2) , −i(2) ]. Let hyi ∈ βe4 (G) be such that y maps onto y¯. If Z(E)y = Z(E)y0 , then this coset describes one of the |CU (¯ y )| = 4 complements to U in hU, Z(E)y0 i, which are all conjugate to hZ(E)y0 i. So assume that Z(E)y 6= Z(E)y0 . Then y cannot be an involution and Z(E)y has order 4. If y is bad for U , then y 2 is an involution in E rZ(E) and so χ(y) = χ(y 2 ) = 0 and yV = [i, −i, 1, −1]. If y represents one of the four good cosets, then |χ(y)|2 = 4 and yV = [±i(2) , 1, −1] or [±1(2) , i, −i]. Hence ³ ´ X 1 1 |CV (γ)| ≤ 5 4(2r2 ) + 8(4r) + 4(r2 + 2r) . 2 2 e4 (G) γ∈β

100

It follows that


P e4 (G) γ∈β5 (G)∪β

|CV (γ)| ≤ 50r2 + 164r, which is less than

r4 = |V | since r ≥ 9 (as p = 3 and 4 | r − 1). Hence there is v ∈ V such that CG (v) = CEZ (v) is an elementary abelian 2-group. ¯ is an irreducible 50 -subgroup of Sp4 (2) ∼ Let next p = 5. Then G = S6 , ¯ and we may assume that G ∼ = O+ (2) ∼ = 32 : D8 [Atlas]. For r = 5 we find, 4

using a random search, v ∈ V such that H = CG (v) is a S4 group. Then the nontrivial H-submodule W of V is the deleted permutation module or its tensor product with the sign character, and H has a regular orbit on W . So let r ≥ 25. We may argue as for (i), but using the character table (for T ). We find v ∈ V such that CG (v) maps isomorphically into a Sylow ¯ Using that the elements in G ¯ of order 4 belong to A6 and 2-subgroup of G. that χ vanishes on the elements of order 4 in T ∼ = 2.A6 we see that v is strongly real for G. Also, CG (v) has a regular orbit on V . Let finally p > 5 (r ≥ 13), and let G = G0 . Write E = E0 ◦ Z4 with ∼ + ¯ E0 ∼ = 21+4 + . Let X0 = NG (E0 ). Then X0 = X0 /ZE = O4 (2). As in the second paragraph of the proof for (i) let S be a Sylow 3-subgroup of X0 , ¯ 0 ), and let Y = NX (S). Then which maps isomorphically onto S¯ = O3 (X 0 ¯ 0 of a maximal Y · ZE = X0 and Y ∩ ZE = Z. Let P be the stabilizer in X ¯ totally singular subspace W1 of U0 = E0 /Z(E0 ). Thus P ∼ = Λ2 (W1 ) : L ∼ ∼ ¯ where W1 is the standard module for L = GL(W1 ) = S3 (Appendix B). Also U0 = W1 ⊕W2 = W1 ⊕W3 where the Wi are all the distinct, but isomorphic, ¯ irreducible L-submodules of U0 ∼ = U (noting that |HomL¯ (U0 /W1 , W2 )| = 2). n ¯ Since H (L, Wi ) = 0 for all n ≥ 1, in view of (A2), there exist a subgroup L ¯ and L-invariant elementary abelian of X0 = NGL(V ) (E0 ) mapping onto L subgroups Ai of E mapping onto the Wi . By Appendix (C1) we may choose L such that L ∼ = GL(Ai ) ∼ = S3 . The fixed spaces CV (Ai ) are 1dimensional, pairwise distinct, and centralized by L0 . Since an involution t in L acts as ±1 on these spaces, we may replace t by tz for the generator z of Z(E0 ), if necessary (and L correspondingly), so that L centralizes at least two of these fixed spaces, say CV (A1 ) and CV (A2 ). We may choose v ∈ CV (A1 ) ⊕ CV (A2 ) outside all CV (Ai ). We claim that NZE (F v) = Z. For otherwise NZE (F v)/Z would be an irreducible L-submodule of U ∼ = U0 and hence one of the Wi . It follows that LAi ⊆ NG0 (F v) and that Ai centralizes v, because L centralizes v and [Ai , L] = Ai . Hence the claim. In particular CZE (v) = 1. Let H = CG (v), ¯ be its isomorphic image in G. ¯ and let H ¯ 6= G. ¯ If H ∼ From the character table we infer that H = A6 , S5 or A5 ,


101

then H is a real group, and H has a regular orbit on V by Theorem 7.2a ¯ is contained in a maximal below. H is a real group in the cases where H subgroup of S6 of type S4 × Z2 or in a solvable subgroup of S5 (as H ⊇ L), and then H has a regular orbit on V (as r ≥ 13). It remains to investigate ¯0 ¯ is contained in a maximal subgroup of type O+ (2) ∼ the case where H =X 4 ¯ ⊆ X ¯ 0 since X ¯ 0 is the unique 3-Sylow normalizer in G ¯0 of S6 . Then H ¯ We assert that H ⊆ L × Z2 , which will complete the proof. containing L. ¯ = O3 (X ¯ 0 ) and O3 (H) ∼ Otherwise O3 (H) = S. Changing notation, if necessary, we may let S = O3 (H). Then H ⊆ Y = NX0 (S) and Y /ZS ∼ = D8 . Since S centralizes v and is faithful on V , by Proposition 1.6a we have a proper decomposition V = CV (S) ⊕ [V, S] of F Y -modules (p 6= 3). This is impossible since Y is irreducible on V , as follows from Theorem 1.8b. For ResYS (V ) is the sum of four nonisomorphic 1-dimensional F S-modules corresponding to an Y -orbit on S ] when 3 | r−1, and of two 2-dimensional modules otherwise. ¤ Proposition 6.6c. Let E ∼ = 21+6 − . There exists a strongly real vector v ∈ V for G such that CG (v) has a regular orbit on V . 0 Proof. O− 6 (2) contains no irreducible 3 -subgroup [Atlas, p. 26]. Hence p > 3 and r ≥ 7 (as 4 - r − 1). For r = 7 by computation with [GAP] (or directly) one finds v ∈ V such that CG0 (v) ∼ = S4 , which is a real group and which obviously has a regular orbit on V (dim F V = 8). So let r ≥ 11 in what follows. ∼ 31+2 : GL2 (3) be a maximal subgroup of X ∼ O− (2) which ¯ = Let Y¯ = + 6 1+2 is irreducible. Let S ∼ = 3+ be a subgroup of X such that S¯ = SE/E = O3 (Y¯ ), and let Y = NX (S). By the Frattini argument Y covers Y¯ and so

is irreducible on U = E/Z(E). Each faithful irreducible representation of S has degree 3 or 6 depending on whether the underlying field of scalars contains a primitive 3rd root of unity or not. Since S is faithful on U (and Z(S) fixed point free), therefore even S is irreducible on U . Note that Y inverts the elements of Z(S). It follows that Y ∩ E = Z(E) and that CX (Z(S)) = Y 0 Z(E) has index 2 in Y , mapping onto the maximal subgroup Y¯ 0 ∼ = GU3 (2) of Ω− 6 (2). Using that GL2 (3) has trivial Schur multiplier we 0 see that Y Z(E) = Y 0 ×Z(E) and that Y 0 is a weak holomorph of S ∼ = 31+2 + (as M(Sp2 (3)) = 1).

102


We show that Y 0 is the standard holomorph of S, that is, Y 0 splits over S. In view of (A2) we find a subgroup L0 of Y 0 such that L0 S = Y 0 and L0 ∩ S = Z(S). We have to exclude that L0 /L00 is cyclic (of order 9). In this case L0 , hence Ω− 6 (2), had an element of order 18, which is false [Atlas, p. 27]. Hence there is a subgroup L ∼ = Sp2 (3) in Y 0 , and Y 0 = LS. Since S is faithful on V and Y inverts the elements of Z(S), there is a faithful irreducible F Y -submodule V1 of V of dimension 6. In fact V1 = [V, Z(S)] and the restriction to CX (Z(S)) = Y 0 × Z(E) of V1 affords the sum of two distinct (absolutely) irreducible characters ϕ, ϕ¯ of degree 3, which are faithful and which are interchanged by Y . Let V0 = CV (Z(S)), so that V = V0 ⊕ V1 is a decomposition of F Y -modules (Proposition 1.6a). Since Y is transitive on (S/S 0 )] (and dim F V0 = 2), it follows from Theorem 1.8b that V0 = CV (S). We assert that L is faithful and irreducible on V0 . Assume there is an element x ∈ L of order 4 which is trivial on V0 . We have ϕ(x) = 1 = ϕ(x) ¯ 0 as ResYL (ϕ) is the sum of two irreducible characters of degrees 1 and 2. It thus follows that χ(x) = 1+1+2 = 4, hence |CU (x)| = 16 = 42 by Theorem 4.4. Now the image x ¯ of x in O− 6 (2) has order 4 and preserves an F4 structure on U , because x ¯ ∈ SU3 (2). Since x ¯ is a unipotent transformation (3) ∼ on U = F4 , it must have two Jordan blocks on U of sizes 1 and 2. But this forces that x ¯2 = 1, a contradiction. Thus x does not centralize V0 . Since L/Z(L) ∼ = A4 has no faithful irreducible F -representation of degree 2, we see that L is faithful and irreducible on V0 , as asserted. We infer that L maps injectively into Y /CY (V0 ). Since both Z(E) and Z(L) act as −1 on V0 , it follows that |CY (V0 ) : S| = 2. Moreover CY (V0 )0 = S as Z(E) centralizes S but Z(L) inverts the elements of S/Z(S). At most 8 points in P1 (V0 ) are fixed by some nonidentity element in L/Z(L) ∼ = A4 , because the elements of order 4 in L do not fix any point (as 4 - r − 1). We conclude that there is v ∈ V0 such that NL (F v) = Z(L) (as r ≥ 11). Since S ⊆ CY (V0 ) is irreducible on U = E/Z(E) and E is irreducible on V , we have NE (F v) = Z(E). It follows that the index |CY (v) : CY (V0 )| ≤ 2 and that CY (v)0 = S. Let H = CX (v). Since H ⊇ S acts irreducibly on U , we have H ∩ E = 1. Thus H maps isomorphically ¯ of X. ¯ onto an irreducible subgroup H By (A9), (A10) NX (F v) maps onto a proper (irreducible) subgroup 3 4 ¯ ∼ of X = O− 6 (2) and so its Sylow 3-subgroups have order |S| = 3 or 3 .


103

In the latter case S ⊂ S0 for some Sylow 3-subgroup S0 of NX (F v). But then S0 ⊆ NX (S) = Y , even S0 ⊆ Y 0 = LS. Since S centralizes V0 , it follows that NL (F v) contains an element of order 3, contradicting the fact that NL (F v) = Z(L). Consequently NX (F v)/CX (v) is a (cyclic) 2-group (of order dividing r − 1). Now |NX (F v) : CX (v)| = 2 as 4 - r − 1, and NX (F v) = CX (v) × Z(E). From Lemma 5.1b we infer that CG0 (v) = CX (v) = H. We shall prove that H ⊆ Y and so H/S has order 2 or 4. ¯ is contained in a conjugate of Y¯ , the Sylow 2Assume H 6⊆ Y . If H 0 subgroups of H are cyclic, because the Sylow 2-subgroups of Y¯ /S¯ ∼ = GL2 (3) are semidihedral and S is a Sylow 3-subgroup of H. By a simple transfer argument H 0 has a normal 2-complement. But then H ⊆ NX (S) = Y . So ¯ being irreducible, is contained in a maximal subgroup P0 = 33 : (S4 ×Z2 ) H, ∼ ¯ ∼ of X = O− 6 (2) = PSp4 (3) [Atlas, p. 26]. Here S4 = PGL2 (3) is irreducible 3 on R = 3 , because P = R : GL2 (3) is a maximal parabolic of Sp4 (3) stabilizing an isotropic line W , and R ∼ = Sym2 (W ). Now S¯ is a Sylow ¯ of order 33 , which by assumption is not normal in H. ¯ We 3-subgroup of H 2 ¯ may assume that |R ∩ S| = 3 (Sylow). No element of order 3 in P0 /R ¯ and so |CH¯ (R ∩ H) ¯ : R ∩ H| ¯ ≤ 2. It follows that centralizes R ∩ S¯ = R ∩ H ¯ H¯ (R∩H) ¯ is a subgroup of GL2 (3) having no normal Sylow 3-subgroups, H/C hence containing SL2 (3). On the other hand, this is a homomorphic image of a subgroup of S4 × Z2 , which is impossible. √ ResX (χ) is rational-valued, because on X the values of χ lie in Q( 2) √H or Q( −2) by Theorem 4.3e, and the exponent of H is 6 or 12. Hence it remains to show that H has a regular orbit on V . There are at most 3 · 33 = 81 involutions in H, each having at most r6 fixed points on V . If y is one of the 2 · 13 elements of order 3 in H, then dim F CV (y) ≤ 4 by P Theorem 6.2c. As r ≥ 11, γ∈β(H) |CV (γ)| ≤ 81r6 + 13r4 < r8 = |V |. ¤ 6.7.

The Remaining Cases

We often shall use that certain groups cannot appear as subgroups of X, e.g., by the nonsplitting properties stated in (A9), (A10). Other groups will be ruled out on the basis of Theorem 4.4 and the character table. For example, let m = 5 and let X be the holomorph to any of 21+10 ±,0 . Assume Y = A9 occurs as a subgroup of X. Then there must be a nonnegative P6 n χ of the first six irreducible charinteger combination ResX Y (χ) = i=1 P i i P acters of A9 [Atlas, p. 37] such that i ni χi (1) = 25 and | i ni χi (y)|2 = 0 or a 2-power (y ∈ Y ). This is impossible.

104


Proposition 6.7a. Let E ∼ for m ≥ 3. There exists a strongly real = 21+2m + vector v ∈ V for G such that CG (v) has a regular orbit on V . Proof. By Theorem 6.3b it suffices to consider the cases m = 3, 4, 5. By Theorem 4.5b there is a subgroup L ∼ = Lm (2) of X normalizing two elementary abelian subgroups A, A∗ of E yielding a decomposition U = W ⊕ W ∗ into totally singular subspaces. Here A ∼ = W is the standard module of L and A∗ ∼ = W ∗ its dual. Recall also that ResX L (χ) = 2·1L +ξ for some rational-valued irreducible (Weil) character ξ. We further know that NX (L) = LZ(E)hτ i where τ interchanges A and A∗ and induces the inverse transpose automorphism on L (τ 2 ∈ Z(E)). Now CV (A) and CV (A∗ ) are distinct 1-dimensional subspaces of V , and L = L0 centralizes Ve = CV (A) ⊕ CV (A∗ ). Also, τ interchanges CV (A) and CV (A∗ ) and fixes no point in P1 (Ve ) if τ 2 6= 1 (as 4 - r − 1) and two points otherwise. We choose v ∈ Ve outside CV (A) and CV (A∗ ) and, if possible, such that v is no eigenvector of τ . Thus if τ stabilizes F v, then r = 3 and τ is an involution. Let H = CG0 (v). Assume H ∩ ZE 6= 1. Since H ∩ Z = 1, H ∩ ZE then is an L-invariant elementary abelian subgroup of ZE whose image in U = ZE/Z is nonzero and totally singular and so must be W or W ∗ , because U = W ⊕ W ∗ and W , W ∗ are irreducible and nonisomorphic L-modules (Lemma 4.1c). Since H contains [H ∩ EZ, L], which is A or A∗ , it follows that A or A∗ centralize v. This contradicts the choice of v. Hence H ⊇ L map injectively onto ¯ ⊇L ¯ of G ¯0 ∼ subgroups H = O+ 2m (2). ¯ contains NG¯ (L) ¯ = hL, ¯ τ¯i. Then there Let τ¯ = τ ZE. Suppose H 0

is an element τ0 ∈ H with τ0 ZE = τ¯, and then L = Lτ0 must agree (within H). Thus τ0 and τ act on L in the same manner. We deduce that τ τ0−1 ∈ CZE (L) = Z. As τ0 centralizes v, τ fixes F v. Thus r = 3 and τ is an involution. ¯ is neither G ¯ 0 = O+ (2) nor Ω+ (2), nor is any subBy (A9), (A10) H 2m 2m ¯ is contained in a maximal group of odd index in these groups by (A4). If H

parabolic P of Ω+ 2m (2) stabilizing a maximal totally singular subspace of ¯ on U , and U , this subspace must be W or W ∗ = W τ , by the action of L ¯ necessarily H = L as the Levi complement L is irreducible on the unipotent radical R ∼ = Λ2 (W ) resp. Λ2 (W ∗ ) of P (Appendix B and Appendix A7). Let m = 3. Then H = L or r = 3 and H = Lhτ i ∼ = PGL2 (7), the ∼ further possibilities H = A7 , S7 being ruled out by Theorem 4.4 and the character tables. The extensions of ξ to PGL2 (7) take only real values


[Atlas, p. 3]. Also, X

105

|CV (γ)| ≤ 21r6 + 28r5 + 28r4 < r8 = |V |

γ∈β(H)

for r > 3. For r = 3, H ∩ G is contained in groups of type D16 (not in L) or 7 : 2 (not in L). An involution of H in L has r6 fixed points on V , an involution outside r5 . Use that 5r6 + 4r5 < r8 respectively 7r5 + r2 < r8 . Let m = 4. We assert that then H = L. We inspect the maximal ∼ subgroups of O+ 8 (2) [Atlas, p. 85]. The possibilities H = A9 , S9 are ruled ¯ cannot be in a maximal subgroup of O+ (2) out by the character tables. H 8 ¯ ⊇ L). ¯ If of type Sp6 (2) × Z2 , the stabilizer of a nonsingular point (as H ¯ 6= L, ¯ then H ¯ must be contained in an (irreducible) Sp6 (2) group [Atlas, H ¯ p. 46], which in turn contains an S8 subgroup. Thus if H 6= L, then τ¯ ∈ H, ∼ ∼ r = 3 and either H = Lhτ i = S8 or H = Sp6 (2). But in this case χ restricts to an S8 subgroup as 2χ1 + χ3 (Atlas notation) where, by Theorem 4.4, χ3 (2C) = −4 on the involutions 2C outside A8 . But then χ3 (10A) = +1 and χ(10A) = 3, contradicting Theorem 4.4. Hence the assertion. The counting method yields a regular orbit on V for H ∩ G for r > 3 (hence r ≥ 7). For r = 3 use that H ∩ G is contained in groups of type 24 : 22 , 23 : 7, 5 : 4 or 5 : 22 [Atlas], and that 26 r12 < r16 . Let m = 5. Then H = L or r = 3 and H = Lhτ i ∼ = L5 (2) : 2 by

inspection of the maximal subgroups of O+ 10 (2) [Atlas, p. 146]. Observe that ξ extends to real-valued characters of L5 (2) : 2 [Atlas, p. 70]. As before we get a regular orbit for H ∩ G. ¤ Proposition 6.7b. Let E ∼ for m ≥ 3. There exists a strongly real = 21+2m 0 vector v ∈ V for G such that CG (v) has a regular orbit on V . Proof. By Theorem 6.3b we only have to examine the cases m = 3, 4, 5. Write E = E0 ◦Z4 where E0 ∼ . Let Y = NX (E0 ). Then Y¯ = Y /E ∼ = 21+2m = + + ∼ ¯ O2m (2) is a maximal subgroup of X = Sp2m (2). As before pick a subgroup L∼ = Lm (2) of Y and L-invariant elementary abelian subgroups A, A∗ of E0 of order 2m . Again ResX L (χ) = 2·1L +ξ for some rational-valued irreducible (Weil) character ξ. We also find τ ∈ X inducing the inverse transpose automorphism on L such that NG0 (L) = LZhτ i and τ 2 ∈ Z(E) (see the remark to Theorem 4.5b). This τ interchanges A = [Ω1 (AZ(E)), L] and A∗ = [Ω1 (A∗ Z(E)), L]. Now L = L0 centralizes Ve = CV (A) ⊕ CV (A∗ ), and τ interchanges the 1-dimensional fixed spaces CV (A) and CV (A∗ ). Multiply τ by a scalar in Z, if necessary, such that its order is as small as possible.

106


Let m = 3. Choose w ∈ Ve outside CV (A) = and CV (A∗ ) and, if possible, such that τ centralizes w. Let H = CG0 (w). As before one shows that H ∩ ZE = 1. So by (A10) H ⊇ L map injectively onto proper ¯ ⊇ L ¯ of G ¯0 ∼ subgroups H = Sp6 (2). We cannot have H ∼ = A7 , S7 by the ∼ character tables and (4.4). Similarly H is not isomorphic to Ω+ 6 (2) = A8 . ¯ is contained in a maximal G2 (2) It remains to study the situation where H ¯ subgroup of G0 [Atlas, p. 46]. We cannot have H ∼ = G2 (2) by the character table and (4.4). Suppose H ∼ = G2 (2)0 . Then H = H 0 ⊆ G00 = X 0 and [Atlas, p. 14] ResX H (χ) = χ1 + χ4 or χ1 + χ5 , where χ4 and χ5 are algebraically conjugate and not real-valued (but fuse in G2 (2); ResX H (χ) = χ1 + χ3 is excluded by considering the value on the conjugacy class 6A). There is a ¯ in G ¯ 0 , namely that normalized by τ¯ = τ ZE. unique G2 (2)0 overgroup of L ¯ ¯ τ i by Theorem 4.5b, and use that (To see this, recall that NG¯ 0 (L) = Lh¯ Sp6 (2) contains a unique conjugacy class of G2 (2)0 subgroups and that each G2 (2)0 group has a unique conjugacy class of (maximal) L3 (2) subgroups.) ¯ (¯ ¯ Hence both H and H τ contain L and map (isomorphically) onto H τ 6∈ H). Now U is the unique (absolutely) irreducible F2 H-module of dimension 6 (arising naturally from the action of G2 (2) on the F2 Cayley algebra [Atlas, p. 14], [B-Atlas, p. 23]). One knows that H1 (H, U ) has order 2 [Sin, 1996]. Using that H ∼ = G2 (2)0 is perfect and has trivial Schur multiplier we see that there are just 2 conjugacy classes of complements to E in HE. If two such complements, say H and H τ , are conjugate −1 under E, say H τ = H y (y ∈ E), then Ly = {x[x, y] | x ∈ L} ⊆ H and [L, y] ⊆ H ∩ E = 1, that is, y ∈ Z(E) ⊆ Z. We conclude that either ¯ H = H τ , or H 6= H τ are the unique overgroups of L mapping onto H. ¯ is irreducible on U = E/Z(E). Hence H = H 0 Clearly hH, Ai ⊇ E as H does not centralize CV (A). It follows that CVe (H) = F w. If τ fixes F w, then by the choices of τ and of w it centralizes w. But τ 6∈ H = CG0 (w). We conclude that τ does not fix F w, whence H 6= H τ , and these two groups fix just the two points F w and F wτ in P1 (Ve ). Note that |P1 (Ve )| ≥ 6. If we choose v ∈ Ve such that F v is different from these two points, different from CV (A) and CV (A∗ ), and if possible such that F v is the eigenspace of τ on Ve to the eigenvalue 1, then H0 = CG0 (v) either is L or τ is an involution and H0 = Lhτ i ∼ = PGL2 (7). This v is strongly real for G, and in the worst case we get the estimate X γ∈β(H0 )

|CV (γ)| ≤ 21r6 + 28r5 + 28r4 .


107

This is less than |V | = r8 except when r = 5. But in this exceptional case H0 , being a 50 -group, has a regular orbit on V by Theorem 7.2a below. For m = 4, 5 we choose v ∈ Ve outside CV (A) and CV (A∗ ) such that F v is not fixed by τ . Then H = CG0 (v) agrees with L, and L ∩ G has a regular orbit on V . In the m = 4 case this follows by inspection of the maximal subgroups of Sp8 (2) [Atlas, p. 123]. In the m = 5 case we deduce from Lemma 6.3a (and Table 3.5.C in [Kleidman–Liebeck, 1990]) that the (isomorphic) image of H in Sp10 (2) must be properly contained in O± 10 (2) subgroups, and inspection of the [Atlas, pp. 146, 147] yields H = L. ¤ Proposition 6.7c. Let E ∼ with m ≥ 4. There exists a strongly = 21+2m − real vector v ∈ V for G such that CG (v) has a regular orbit on V . Proof. By Theorem 6.3b we only have to consider the cases m = 4, 5. 1+2(m−1) Decompose E = E0 ◦ E1 where E0 ∼ and E1 ∼ = 2+ = Q8 . Let X0 = CX (E1 ) and X1 = CX (E0 ). Then X0 is the standard holomorph of E0 (in F ) and X1 that of E1 (in F ). Moreover, the central product Y = X0 ◦ X1 (amalgamating Z(E0 ) = Z(E1 )) is a maximal subgroup of X. As an Y module we may decompose V = V0 ⊗F V1 where V0 is the natural module for X0 and V1 that for X1 . Recall that X1 ∼ = 2+ S4 = GL2 (3) if r ≡ 3 (mod 8) − and X1 ∼ = 2 S4 if r ≡ 7 (mod 8) (4 - r − 1). Let χ0 and χ1 denote the obvious characters of X0 and X1 , respectively, so that ResX Y (χ) = χ0 ⊗ χ1 . ∗ ∗ ∼ Let L0 = Lm−1 (2), A0 , A and W0 , W be defined for X0 as in Theorem 0

0

4.5b. Hence U0 = E0 Z/Z = W0 ⊕ W0∗ where W0 is the standard module for L0 and W0∗ is its nonisomorphic dual. Here Ve0 = CV0 (A0 ) ⊕ CV0 (A∗0 ) is a 2-dimensional subspace of V0 which is centralized by L0 = L00 . Further NX0 (L0 ) = L0 Z(E0 )hτ0 i for some τ0 which induces the inverse transpose 0 automorphism on L0 (τ02 ∈ Z(E0 )). We know that ResX L0 (χ0 ) = 2 · 1L0 + ξ0 , where the Weil character ξ0 is rational-valued. ∼ S3 , and choose v0 ∈ Ve0 For r = 3 pick any v1 ∈ V ] , so that CX (v1 ) = 1

1

outside CVe (A0 ) ∪ CVe (A∗0 ), and let v = v0 ⊗ v1 . Otherwise r ≥ 7, and we 0 0 choose v ∈ Ve = Ve0 ⊗F V1 which is not centralized by A0 , A∗0 nor by any element of prime order (2 or 3) in hτ0 iX1 Z. (We have at most 12 subgroups of order 3 and at most 2 + 2 · 12 involutions in hτ0 iX1 Z outside Z, each with at most r2 fixed points on Ve , and 38r2 < r4 = |Ve |.) Let L = L0 × CX1 (v1 ) for r = 3 and L = L0 otherwise. Of course χ takes only rational values on L. In view of the m = 3, m = 4 cases of the proof for Proposition 6.7a it is clear that there there is a regular (L ∩ G)-orbit on V .

108


Let H = CG0 (v), so that H ⊇ L. We assert that H ∩ ZE = 1. Otherwise H ∩ ZE is an L0 -invariant elementary abelian subgroup of ZE mapping onto an L0 -submodule 6= 0 of U = U0 ⊥U1 = (W0 ⊕ W0∗ )⊥U1 . U is a completely reducible F2 L0 -module, W0 , W0∗ being irreducible and U1 centralized by L0 , each point in U1 being nonsingular. Hence the image of H ∩ ZE must be W or W ∗ . But this implies that one of A or A∗ centralizes v, in contrast to the choice of v. Therefore H ⊇ L map isomorphically onto ¯ ⊇L ¯ of G ¯0 ∼ (proper) subgroups H = O− 2m (2). ∼ O+ ¯0 = ¯ when r > 3. The element τ¯0 = τ0 ZE of X (2) is not in H 2(m−1)

For otherwise H contains an element τ with τ ZE = τ¯0 , both L = L0 and ¯=L ¯ τ¯0 . Thus L0 = Lτ and both Lτ being subgroups of H mapping onto L 0 τ0 , τ induce the inverse transpose automorphism on L0 . We conclude that τ · τ0−1 ∈ CZE (L0 ) = ZE1 and so τ = τ0 x1 z ∈ H for some x1 ∈ E1 and z ∈ Z, against our construction. ∼ O− (2) [Atlas, ¯0 = Let m = 4. Inspection of the maximal subgroups of G 8 ¯ must be contained in subgroups of the following types: p. 89] shows that H ¯ 0 -conjugate of Y¯ = Y Z/ZE, isomorphic to O+ (2) × O− (2) ∼ (i) A G = 6

2

S8 × S3 . (ii) A maximal parabolic stabilizing a maximal totally singular subspace of U . (iii) The stabilizer Sp6 (2) × Z2 of some nonsingular point in U . ¯ is contained in an irreHere we already ruled out the possibility that H − ∼ PSL2 (7) would be ¯ = ducible PGL2 (7) subgroup of O8 (2). In this case L ∼ ¯ normal in H = PGL2 (7) and act on U with irreducible constituents of ¯ ⊆ Y¯ by the action of equal dimensions (Theorem 1.8). In (i) we have H ¯ L on U and, as in the proof for Proposition 6.7a, we get H = L or r = 3 ∼ PGL2 (7) × S3 . In (ii) the parabolic P is the stabilizer of W0 or and H = ¯=L ¯ τ¯0 . So P = R : L, ¯ where the unipotent radical W ∗ = W τ¯0 , and P ⊇ L 0

0

R is a central extension of W0 ⊗ U1 by Λ2 (W0 ) if P = NG¯ 0 (W0 ) (Appendix ¯ cannot contain Λ2 (W0 ) (resp. Λ2 (W ∗ )) for otherwise H contains B2). H 0 an overgroup of L0 mapping isomorphically onto a maximal parabolic of 2 Ω+ 6 (2), contradicting (A10). Using that L0 is irreducible on Λ (W0 ) (resp. 2 ∗ Λ (W0 )) we get H = L. In (iii) the nonsingular point must be in U1 , hence ¯ is contained in an Sp6 (2) group by the choice of v, because r > 3. Here H


109

the direct factor Z2 is such that it preserves U1 and centralizes U0 = U1⊥ . + ¯ ∼ We cannot have H = Sp6 (2) or O+ 6 (2), Ω6 (2) by (A10). By inspection of the maximal subgroups of Sp6 (2) and G2 (2) [Atlas, pp. 46, 14], and using ¯ we obtain that either H = L or H ∼ that τ¯0 6∈ H, = G2 (2)0 . In the latter case H ⊆ X and ResX H (χ) = 2χ1 + χ4 + χ5 in the notation of the Atlas (p. 14), where χ4 and χ5 are algebraically conjugate and fuse in G2 (2) to a rational-valued character. Also H ∩ G has a regular orbit on V in this case, because X

|CV (γ)| ≤ 63r12 + 336r8 + (28 + 288)r4 < r16 = |V |.

γ∈β(H)

¯0 ∼ Let m = 5. Consider the maximal subgroups of G = O− 10 (2) [Atlas, p. 147]. Recall that A9 does not occur as a subgroup of G0 . Furthermore, − − + ¯ O+ 6 (2) × O4 (2) and O6 (2) × O4 (2) cannot be overgroups of H (by their action on U ), nor M12 : 2 (missing in the [Atlas]) and GU5 (2) : 2 (as they do not have L4 (2) subgroups). It remains to consider cases (i), (ii), (iii) defined as in the m = 4 case. This leads to H = L or r = 3 and H is of type S8 ×S3 or Sp6 (2)×S3 , which are ruled out as in the m = 4 case of the proof ¯ is contained in a stabilizer Sp8 (2) × Z2 of for Proposition 6.7a. In (iii) H a nonsingular point, necessarily in U1 , which implies that r > 3. It follows that either H = L = L0 or H ∼ = Sp2 (6) maps onto on irreducible subgroup ¯ of O+ (2) [Atlas]. However, then H ¯ contains L ¯ 0 h¯ H τ0 i ∼ = S8 , which has also 8 been ruled out. ¤ This completes the proof for Theorem 6.1. At this stage the k(GV ) conjecture is settled for solvable groups G and modules V in characteristic p 6= 3, 5, 7, 13 (by virtue of Theorems 5.2b and 5.4). Moreover, if G is a (solvable) group with a normal 2-complement, the conjecture holds. For the property of having a normal 2-complement is preserved when passing to subgroups, quotient groups and central extensions, so that Theorem 5.4 applies. In particular, the k(GV ) conjecture holds for supersolvable groups [Knörr, 1984], which have a normal 2-complement, and for groups of odd order [Gluck, 1984].

Chapter 7

Reduced Pairs of Quasisimple Type The main task of this chapter will be to determine the reduced pairs (G, V ) of quasisimple type where there is no regular G-orbit on V . The classification of these pairs has been achieved by the efforts of [Liebeck, 1996], [Goodwin, 2000], [Riese, 2001] and [Köhler and Pahlings, 2001]. From this it is easy to classify the nonreal reduced pairs of quasisimple type. 7.1. Nonreal Reduced Pairs Throughout (G, V ) will be a reduced pair over F = Fr with quasisimple core, E. By coprimeness, and by the Feit–Thompson theorem, r is a power of some odd prime p. Embed E and G into GL(V ) and let G0 = NGL(V ) (E). Again we distinguish these pairs by their cores. It turns out that, in this sense, there are just 3 types of nonreal reduced pairs with quasisimple cores. Here all characters involved are Weil characters and, up to three exceptions in type (Sp4 (3)), the pairs are large. Type (2.A5 ) : Let r = p be one of the primes 11, 19, 29 or 59. We have seen in Sec. 3.2 that then GL2 (p) contains a subgroup E ∼ = 2.A5 such (2) that G0 = NGL2 (p) (E) acts transitively on the nonzero vectors in V = Fp . Here G = G0 ∼ = E ◦ Zp−1 is a p0 -group, E is absolutely irreducible on V , and the centralizer in G of a nonzero vector is cyclic of order 5, 3, 2 or 1, respectively. We conclude, in view of Lemma 4.1c, that (G, V ) is a nonreal reduced pair for r = 11 and for r = 19. The Brauer character of E afforded by V is one of the two faithful irreducible Weil characters (of degree 2) of SL2 (5), which are algebraically conjugate and fuse in 2.A5 .2. There are other primes p ≥ 11 where GL2 (p) contains a copy E of 2.A5 = SL2 (5), because this only requires that 5 is a square mod p. We get a further nonreal reduced pair (G, V ) for p = 31, where G ∼ = 2.A5 × Z15 has two orbits on nonzero vectors with point stabilizers of order 3 and 5. Type (2.A6 ) : The group E = 2.A6 has two faithful irreducible characters, χ8 and χ9 , of degree 4 [Atlas, p. 5]. These are Weil characters for E ∼ = 110

Reduced Pairs of Quasisimple Type

111

Sp2 (9), and they fuse in 2.A6 .22 . There is a nonreal extension of χ8 to 2− S6 (not given in the Atlas), and a nonreal extension of χ9 to the isoclinic variant 2+ S6 , each requiring a square root of −3. √ √ Let r = p = 7. Then −3 is in F = F7 but 3 is not. Let Vi be an F E-module affording χi , embed E into GL(Vi ) and let Gi = NGL(Vi ) (E) (i = 8, 9). Then Gi is isomorphic to (2− S6 ) × Z3 resp. to (2+ S6 ) × Z3 . There are three orbits of Gi on Vi] , the point stabilizers being of type Z3 , Z6 and Z3 wr Z2 ∼ = Z3 × S3 (in both cases). This has been proved in Example 5.7b, where we also showed that no vector in Vi is real for Gi . We have two nonisomorphic nonreal reduced pairs (Gi , Vi ) where, however, the underlying groups are isoclinic and the modules behave similarly. Type (Sp4 (3)) : The group E = Sp4 (3) has two faithful irreducible (Weil) characters χ = χ21 or χ22 of degree 4 which are algebraically conjugate and which fuse in Sp4 (3).2 [Atlas, p. 27]. So the resulting reduced pairs are isomorphic. There are unique conjugacy classes of subgroups S ∼ = SL2 (3) 1+2 ∼ and T = 3+ : Q8 of E on which χ splits off the 1-character, and no proper overgroups have this property [Atlas]; Z(S) is generated by an element in class 2A and Z(T ) by one in classes 3A0 B0 . By considering the elements of E in the classes 6E, and their powers, one gets ResE S (χ) = 1S +λ+ξ2√where λ 6= 1S is linear and ξ2 is an irreducible Weil character (Q(ξ2 ) = Q( −3)). Let F = Fr for r = p = 7, 13 or 19, and let V be a (coprime, faithful) F E-module affording χ as a Brauer character. Embed E into GL(V ) through χ, and let G = NGL(V ) (E) = E ◦ Zr−1 . For r = 7 we have two G-orbits on V ] with point stabilizers S.3 = S × Z3 and T.3 = T × Z3 (where A.B indicates that A is the point stabilizer taken in E). For r = 13 we have four G-orbits with stabilizers S.3 = S × Z3 , T.3 = 31+2 : Sp2 (3), + Q8 .6 = SL2 (3) ◦ Z4 and Z3 .6 = Z3 wr Z2 . For r = 19 there are five Gorbits with stabilizers S.3 = S × Z3 , T.3 = 31+2 : Sp2 (3), Q8 .3 = SL2 (3), + (2) Z3 .3 = Z3 and 1.Z9 . In each case (G, V ) is nonreal reduced, as it is (E, V ) for r = 7 and (E ◦ Z6 , V ) for r = 13, 19. Theorem 7.1. Up to isomorphism there are exactly 11 nonreal reduced pairs of quasisimple type, and these are described above. All other reduced pairs (G, V ) of quasisimple type admit a real vector v ∈ V such that CG (v) has a regular orbit on V . This theorem will be derived from the classification of the reduced pairs of quasisimple type admitting no regular orbits.

112


7.2. Regular Orbits We have seen in Example 5.1a that if (G, V ) is a “permutation pair” over F of degree d and r = p is equal to d + 2 or d + 3, then there is no regular G-orbit on V . These reduced pairs admit always strongly real vectors, and regular vectors if r ≥ d + 4. Theorem 7.2a. Suppose the pair (G, V ) does not admit a regular orbit. Then G0 = NGL(V ) (E) is a p 0 -group. Up to a few isoclinisms, the isomorphism class of (G0 , V ) is determined by G0 , d = dim F V , and possible r = |F |. Excluding the isomorphism classes of the permutation pairs these are listed below, together with the isomorphism type of a point stabilizer H = CG0 (v) of smallest possible order (depending on r): G0 A5 × Z 2.A5 ◦ Z 3.A6 ◦ Z 2.S6 ◦ Z 2.A7 ◦ Z L2 (7) × Z Sp4 (3) ◦ Z PSp4 (3) × Z PSp4 (3).2 × Z Sp6 (2) × Z (U3 (3) × Z).2 U3 (3) × Z U3 (3).2 × Z 61 .U4 (3).22 ◦ Z (U5 (2) × Z).2 2.O+ 8 (2) ◦ Z 2.J2 ◦ Z

d 3 2 3 4 4 3 4 5 6 7 6 7 7 6 10 8 6

r 11 11, 19, 29, 31, 41, 49, 61 19, 31 7 11 11 7, 13, 19, 31, 37 7, 13, 19 7, 11, 19 11, 13, 17, 19 5 5 5 13, 19, 31, 37 7 11, 13, 17, 19, 23 11

minimal stabilizer(s) Z2 Z5 , Z3 , Z2 , Z3 , Z2 , Z2 , Z2 Z2 , Z2 Z3 Z3 Z2 H1 , H2 , Z9 , Z3 , Z2 S4 , V4 , Z2 D12 , V4 , Z2 (3) Z2 , V4 , Z2 , Z2 S3 Z2 Z2 S4 × Z2 , S3 × Z2 , V4 , Z2 V4 S 4 × Z 2 , S 4 , S 3 , V4 , Z 2 S3

Remarks. There are six isoclinic but not isomorphic pairs in the above list (2± S6 , the unitary groups 61 .U4 (3).22 ◦Z for r = 19, 31, and the orthogonal groups for r = 11, 19, 23). For the groups Sp4 (3) ◦ Z the point stabilizers H1 = SL2 (3) × Z3 (r = 7) and H2 = Z3 wr Z2 (r = 13) already appear in Sec. 7.1, and these are the only nonreal reduced pairs where there is no cyclic point stabilizer. It is easy to determine the possible subgroups G of G0 for which there is no regular orbit, too.


113

Except for G0 = 2.A5 ◦ Z, d = 2, r = 49 always r = p is a prime. Here we√deal with the two Weil characters of E = 2.A5 of degree 2 , which require 5 and which fuse in 2.S5 . Their sum is the Brauer character of an irreducible (but not absolutely irreducible) F7 E-module Ve which extends ∼ e =N e e to G e ) (E) = (2.A5 ◦ Z48 ).Z2 . There is no regular G-orbit on V GL(V ∼ (minimal stabilizer H = V4 ). There is only one further example of this kind: E ∼ = L2 (7) has a faithful irreducible module Ve of degree 6 over F = F5 which, as before, is not absolutely irreducible (affording χ2 + χ3 in the Atlas notation, p. 3). ∼ e=N This extends to G e ) (E) = (L2 (7)×Z24 ).Z2 , and there is no regular GL(V e G-orbit on Ve (minimal stabilizer H ∼ = Z2 ). The proof for Theorem 7.2a will be carried out in Secs. 7.3 till 7.10. We now state some consequences. Corollary 7.2b. The reduced pairs described in Sec. 7.1 are the unique ones, up to isomorphism, which are nonreal of quasisimple type. Proof. It suffices to consider the reduced pairs listed in Theorem 7.2a. If the minimal stabilizer H ∼ = CG0 (v) is a real group, the result follows. It therefore remains to show that real vectors exist in the following two cases: Case 1: G = 2.A7 × Z5 , d = 4, F = F11 Let E = 2.A7 . Note that χ = χ10 resp. χ11 in the notation of the Atlas (p. 10). Straightforward computation shows that there is v ∈ V such that H = CE (v) is generated by an element of order 3 in the conjugacy class 3B, where χ takes the value 1. The normalizer NE (H) is a 50 -group. Hence CG (v) = CE (v) by (5.1b) and so v is a strongly real vector for G. Case 2: G = Sp4 (3) × Z15 , d = 4, F = F31 We prove that there is a real vector v ∈ V for G with CG (v) ∼ = Z5 . (There is no strongly real vector in this case). Let E = Sp4 (3). Note that χ = χ21 resp. χ22 in the notation of the [Atlas, p. 27]. We first show that there is a regular E-orbit on V = Vχ . The elements of order 4 and 5 in E do not have nonzero fixed points on V . Each of the 45 noncentral involutions in E has r2 fixed points, the 2 · 40 elements belonging to the classes 3A0 B0 have r fixed points each, and the 2 · 240 elements belonging to 3D0 have r2 fixed points on V . Now use that 45r2 + 40r + 240r2 < r4 = |V | (as r = 31). So let v ∈ V ] belong to a regular E-orbit. Let y ∈ E be an

114


element in the (unique) conjugacy class 5A0 of E of elements of order 5, where χ takes the value −1. Then v, vy, vy 2 , vy 3 , vy 4 are linearly dependent (d = 4). Hence vy = vz for some z ∈ Z = Z30 of order 5. It follows that H = CG (v) contains z −1 y and that NE (F v) contains y. But CE (y) = hyi and NE (F v) is cyclic (isomorphic to a subgroup of F ? ∼ = Z30 ). We conclude that NE (F v) = hyi and, in view of Lemma 5.1b, that H = hz −1 yi. We 2 3 4 have (z −1 y)V = [1, z 2 , z 3 , z 4 ] and so ResG H (χ) affords 1H + λ + λ + λ for 2 some linear character λ of order 5. The F H-module affording 1H + λ + λ3 is self-dual, faithful, and is a submodule of ResG ¤ H (V ), as desired. Corollary 7.2c. Suppose (G, V ) is a reduced pair of quasisimple type having no regular orbit on V . Then there is v ∈ V such that CG (v) has a regular orbit on V , and v can be chosen to be real for G unless (G, V ) is nonreal reduced. Whenever H is a subgroup of G0 satisfying H ∩ Z = 1 which is isomorphic to a minimal stabilizer as given in Theorem 7.2a, then H has a regular orbit on P1 (V ). Proof. From Eq. (3.4b) it follows that if H is abelian with H ∩ Z = 1, then H has a regular orbit on P1 (V ). If (G, V ) is a permutation pair (of degree d), by Example 5.1a there exists v ∈ V such that H = CG (v) ∼ = Sd and ResG (V ) is the permutation module for S over F , which has a regular d H orbit on V . The assertion also holds for the two pairs studied in Corollary 7.2b. The minimal stabilizers H appearing in Theorem 7.2a which are real groups, are easily treated (with the help of the character tables). Hence it remains to examine the nonreal reduced pair of type (Sp4 (3)) for r = 7, 13, where H = H1 or H2 . So let E = Sp4 (3) and χ = χ21 resp. χ22 [Atlas, p. 27]. Consider first the situation where r = 7 and H ∼ = SL2 (3) × Z3 . The unique central involution in H must map into the class 2A of EZ/Z ∼ = U4 (2) and so fixes just 2(r + 1) = 16 points in P1 (V ) (since χ(2A0 ) = 0). There are four conjugacy classes of elements of order 3 in E. The elements in the classes 3A0 B0 have spectral pattern [ε(3) , 1] for some primitive 3rd root of unity ε, hence fix (r3 − 1)/(r − 1) + 1 = 58 points in P1 (V ). Elements in the class 3C0 fix 2(r + 1) = 16 points (since χ(3C0 ) = −2), and those in 3D0 just (r + 1) + 1 + 1 = 10 points (as χ(3D0 ) = 1). We assume the worst case that all elements of order 3 in H belong to 3A0 B0 . Let Y be a noncentral subgroup in H of order 3. Then the four H-conjugates of Y generate a subgroup S ∼ = SL2 (3) of H, and the subgroups in S of prime order fix at most 16 + 4 · 58 = 248 points in P1 (V ). Noting that |P1 (V )| = (r4 − 1)/(r − 1) = 400, there are thus


115

at least 400 − 248 = 152 points which belong to regular orbits of S on P1 (V ). Hence there are at least 7 regular S-orbits on P1 (V ), including a total number of 7 · |S| = 168 points. There are three (central) subgroups (of order 3) in H outside S. By common diagonalization the intersections of the 3-dimensional eigenspaces of their generators on V have dimension at least 2. Hence the three subgroups of H outside S leave at most 58 + 2 · 50 = 158 points in P1 (V ) invariant. There is a point which is not fixed by any of these subgroups and belongs to a regular orbit of S, as desired. In the r = 13 case H ∼ = Z3 wr S2 has four subgroups of order 3 and three subgroups of order 2, each fixing at most 1 + (r2 + r + 1) points in P1 (V ). The result follows. ¤ 7.3. Covering Numbers, Projective Marks For convenience we assume that the pair (G, V ) is large (Z ⊆ G). Recall that its core E = E 0 is perfect, L = E/Z(E) is nonabelian simple and that Z(E) is part of the Schur multiplier M(L). So E is an epimorphic image of the unique (universal) covering group of L (Appendix A). Usually M(L) is “small”, and is cyclic in most cases. At any rate, Z(E) is a cyclic ¯ = G/Z and G ¯ 0 = G0 /Z are subgroups of part of M(L). The groups G ¯ and G ¯ 0 are almost Aut(L) containing L (as a normal subgroup). So G simple groups, and we say that G, G0 are almost quasisimple (just meaning that they are almost simple over their centre). The group of (outer) automorphisms Out(L) of L is known to be solvable (Schreier conjecture), and usually it is “small” as well. For example, if L is one of the 26 sporadic simple groups, then M(L) is cyclic and Out(L) has order 1 or 2, and if it has order 2, then it inverts the elements in M(L). We associate to (G, V ) the Brauer character χ of G afforded by V , which may be understood as an ordinary (absolutely) irreducible character of E extended to G. So we are interested in the faithful irreducible characters of E or, more generally, in the projective irreducible representations of L. The character χ may be regarded as afforded by a faithful projective ¯ By Schur every projective C-representation of G ¯ can representation of G. be lifted to an ordinary representation of its Schur cover. In order to prove Theorem 7.2a we have to work through the classification of the finite simple groups. Usually we proceed in two steps: We use counting arguments as described in Sec. 5.6, as well as lower bounds for d = dim F (V ), to reduce the discussion to “small” simple groups L and

116


“small” degrees d, and “small” r. It turns out that in most of the remaining cases G = G0 = X ◦ Z where either X = E.a is an Atlas groups or where χ is a Weil character. Whenever this does not suffice to finish a proof, we use computational methods following [Köhler–Pahlings, 2001]. ¯ be the smallest integer d0 > 1 such that there is a projective Let R0 (G) ¯ of degree d0 in characteristic 0. Since G ¯ irreducible representation of G 0 ¯ = Rp (G) ¯ (similarly defined), hence d ≥ d0 . Lower is a p -group, R0 (G) ¯ ¯ is bounds for R0 (G) are given in [Schur, 1911] and [Rasala, 1977] when G ¯ an alternating or symmetric group, and in [Tiep–Zalesskii, 1996] when G is a simple classical group (Appendix C), for all simple groups of Lie type by [Seitz–Zalesskii, 1993]. For linear groups GLm (q), symplectic groups Sp2m (q), q odd, and unitary groups SUm (q) this minimal degree is, with few exceptions, attained by the Weil characters (Appendix C and Theorems 4.5a, 4.5b). Being able to treat these Weil characters one can proceed to the characters of the second and third lowest degrees. We usually begin by applying the following crude estimate, already used in [Liebeck, 1996]. For any nontrivial element x in the automorphism group of the simple nonabelian group L let us define the covering number c(x) as the minimum number of L-conjugates of x required to generate the subgroup hL, xi of Aut(L). (This is not common terminology!) Let then (7.3a)

c(L) = max{c(x) : 1 6= x ∈ Aut(L)}.

Upper bounds for c(L) are available (see below). Lemma 7.3b (Liebeck). The fixed point ratio f (g, V ) ≤ 1 − 1/c(Zg) for any g ∈ G r Z. If there is no regular G-orbit on V , then ¯ c(L)/d ≤ (2|Aut(L)|)c(L)/R0 (L) . r ≤ (2|G|)

Proof. Write c(g) = c(Zg). Since the [V, g]x = [V, g x ], x ∈ E, generate V = [V, E], by definition dim F [V, g] ≥ d/c(g). By Proposition 1.6a, V = [V, g] ⊕ CV (g). Hence dim F CV (g) ≤ d − d/c(g). Of course c(L) ≥ c(g) ≥ 2. P By (5.6d) z∈Z |CV (zg)| ≤ 2rd−d/c(g) . If there is no regular G-orbit on V , therefore ¯ d−d/c(L) . rd ≤ 2|G|r This gives the final statement.

¤


117

Now suppose G = X ◦ Z where X = n.L.a is an Atlas group (and ¯ with the almost Z(X) = Z ∩ X has order n ). Then we may identify G s ¯ ¯ simple group X = X/Z(X). Choose a set {Xi }i=1 of representatives for ¯ and let Xi be any (small) the conjugacy classes of the subgroups of X, ¯ i . Define mij as the number of fixed points subgroup of X mapping onto X ¯ j in its action on the transitive X-set ¯ ¯ i \X ¯ of right cosets mod X ¯i. of X X ¯ ¯ The isomorphism type of Xi \X is determined by the ith row vector of the ¯ i | ≤ |X ¯ j | for i ≤ j table of marks (mij ). Arranging matters such that |X ¯ ¯ this is a lower diagonal matrix with mii = |NX¯ (Xi ) : Xi | on the diagonal. Proposition 7.3c. Let t be any positive integer divisible by n = |Z(X)| and dividing exp (X). For each i there exists a distinguished rational polynomial µtχ,X¯ i of degree less than d = χ(1), with the following property: Whenever r is a power of a prime not dividing |X| such that t | r − 1 and χ can be written in F = Fr , afforded by V = Vχ , then Gt = X ◦ Zt has exactly r−1 t ] ¯ ¯ i (r) orbits on V admitting a point stabilizer mapping onto Xi . t · µχ,X Proof. The crucial point will be the computation of these polynomials (their existence being somehow evident). Given r we usually regard the µtχ,X¯ i as polynomials in r (omitting the argument). Of course, Zt denotes the subgroup of order t of Z = Zr−1 and Gt is the central product over Z(X). If for instance µtχ,1 vanishes (at r), there is no regular Gt -orbit on V , hence no regular orbit for G = X ◦ Z. If µtχ,1 does not vanish and t is a proper divisor of r(X) = gcd(r − 1, exp (X)), we have to consider, finally, r(X) the polynomial µχ,1 . Sometimes it happens that CGt (v) = CG (v) for each v belonging to a regular orbit for Gt , that is, these orbits fuse to regular r(X) orbits for G. Then µtχ,1 = µχ,1 , so the polynomial only depends on t. ¯ t = Gt /Zt with X. ¯ Recall Let us construct the polynomials. Identify G that CGt (v) ∩ Zt = 1 and CGt (v)Zt /Zt = CG¯ t (Zt v) for each nonzero vector ¯ v ∈ V , so that the Gt -orbits on V ] are in 1-1 correspondence with the Xorbits on Ω = {Zt v| v ∈ V ] }. (We have Ω = P1 (V ) when Zt = Z.) As an S ¯ ¯ i \X) ¯ for certain multiplicities µi , counting the number X-set, Ω∼ = i µi ·(X ¯ ¯ i . Then |CΩ (X ¯ j )| = P µi · mij of X-orbits with stabilizer conjugate to X i for each j. It is easy to compute these multiplicities once the row vector ¯ 1 )|, · · ·, |CΩ (X ¯ t )|) of marks of X ¯ on Ω is known, and this is given by (|CΩ (X X hχ,λi Xj ¯ j )| = |CΩ (Xj )| = 1 (r − 1) |CΩ (X t λ

for all j. Here the sum is taken over all linear characters λ of Xj of order dividing t (and dividing exp (X)). This is a polynomial in r of degree at most

118


d. Solving (uniquely) the system of linear equations (µ1 , · · ·, µs )(mij ) = (|CΩ (X1 )|, · · ·, |CΩ (Xs )|) we obtain these multiplicities µi . Now, observ¯ i )| = t · |CΩ (X ¯ i )|, we define µt ¯ = t · µi (as a ing that |CP1 (V ) (X r−1 r−1 χ,Xi ¤ polynomial in r). ¯ = X/Z(X) of interest for us the table of For almost all Atlas groups X marks is given in the [GAP] library, and these tables contain generators ¯ i which enable one to compute the character tables of the Xi for the X ¯ i )|. On this basis Köhler has and hence the projective marks |CP1 (V ) (X implemented an algorithm for the computation of the polynomials [Köhler, 1999]. We shall also refer to the results given in [Köhler–Pahlings, 2001]. ¯ Then X ¯ 1 = 1, X ¯2 ∼ ¯3 ∼ ¯4 ∼ Example 7.3d. Let L = A5 = X. = Z2 , X = Z3 , X = ¯5 ∼ ¯6 ∼ ¯7 ∼ ¯8 ∼ ¯9 ∼ V4 , X = Z5 , X = S3 , X = D10 , X = A4 , X = A5 is a set of repre¯ The table of marks is given sentatives for nonconjugate subgroups of X. by   60  30 2     20 0 2     15 3 0 3    (mij ) =  12 0 0 0 2 .    10 2 1 0 0 1     6 2 0 0 1 0 1    5 1 2 1 0 0 0 1 1 1 1 1 1 1 1 1 1 Let us consider one of the (conjugate) Weil characters, χ, of degree 2 for E = 2.L = SL2 (5) = X (characters χ6 , χ7 in [Atlas, p. 2]). Here exp (X) = 60 and G = G0 = X ◦ Z. Let V = Vχ over F = Fr (with r a power of a prime p ≥ 7, and F containing a square root of 5). Given any even, positive divisor t of exp (X) we compute the polynomials µtχ,X¯ i by picking r such that t = r(X) = gcd(r − 1, exp (X)). Thus ¯ = X ¯ = A5 on Ω = P1 (V ). For r(X) = 2, we consider the action of G the corresponding mark vector is (r + 1, 0, 0, 0, 0, 0, 0, 0, 0). This vector is (r + 1, 2, 0, 0, 0, 0, 0, 0, 0) for r(X) = 4, and it is (r + 1, 0, 2, 0, 0, 0, 0, 0, 0) for r(X) = 6. The mark vector is (r + 1, 2, 0, 0, 0, 0, 0, 0, 0) for r(X) = 10, (r + 1, 2, 2, 0, 0, 0, 0, 0, 0) for r(X) = 12, and (r + 1, 2, 0, 0, 2, 0, 0, 0, 0) for r(X) = 20. It is (r+1, 0, 2, 0, 2, 0, 0, 0, 0) for r(X) = 30, and (r+1, 2, 2, 0, 2, 0, 0, 0, 0) for r(X) = 60. We obtain the following polynomials (in r):


µ2χ,1 = (r + 1)/60,

119

µ2χ,X¯ i = 0 for i > 1,

µ4χ,1 = (r − 29)/60, µ4χ,X¯ 2 = 1, µ4χ,X¯ i = 0 for i > 2, µ6χ,1 = (r − 19)/60, µ6χ,X¯ 3 = 1, µ6χ,X¯ i = 0 otherwise, 10 10 µ10 ¯ 5 = 1, µχ,X ¯ i = 0 otherwise, χ,1 = (r − 11)/60, µχ,X 12 12 12 µ12 ¯ 2 = 1 = µχ,X ¯ 3 , µχ,X ¯ i = 0 otherwise, χ,1 = (r − 49)/60, µχ,X 20 20 20 µ20 ¯ 2 = 1 = µχ,X ¯ 5 , µχ,X ¯ i = 0 otherwise, χ,1 = (r − 41)/60, µχ,X 30 30 30 µ30 ¯ 3 = 1 = µχ,X ¯ 5 , µχ,X ¯ i = 0 otherwise, χ,1 = (r − 31)/60, µχ,X 60 µ60 ¯ i = 1 for i = 2, 3, 5 and 0 otherwise. χ,1 = (r − 61)/60, µχ,X

Hence there is a regular G-orbit on V ] if and only if r is different from 11, 19, 29, 31, 41, 49, 61. One also can read off the minimal stabilizers listed in Theorem 7.2a. There is no real vector in V for G if r ∈ {11, 19, 31}, because then there are only stabilizers of order 3 or 5. For r = 49 observe that gcd(r − 1, 60) = 12, in which case we have four G-orbits on V ] with 6 point stabilizers of order 2 (and r−1 6 ·µχ,1 (49) = 4 regular orbits for X ◦Z6 ). 7.4. Sporadic Groups Let L be a sporadic simple group. Then the (abbreviated) character table of L and all possible groups n.L.a can be found in the [Atlas]. Using the information given in the Atlas one can bound c(L). It turns out that c(L) ≤ 6 except possibly for the Fischer groups (c(F i22 ) ≤ 8, c(F i23 ) ≤ 7, c(F i024 ) ≤ 12). In view of Lemma 7.3b this reduces the discussion to some few groups L admitting faithful projective representations of small degrees. Let L = M11 be the smallest Mathieu group. Then L = E and G = G0 = E × Z. Consider the character χ = χ2 of degree 10 [Atlas, p. 18]. Let x be one of the |2A| = 165 (conjugate) involutions in L. From χ(x) = 2 we infer that xV = [1(6) , −1(4) ]. Let y be one of the |3A| = 2 · 220 (conjugate) (3) (3) elements in L of order 3. From χ(y) = 1 we infer that yV = [1(4) , z1 , z2 ] where the zi are the primitive 3rd roots of unity (provided 3 | r − 1). Let u be one of the |5A| = 4 · 396 (conjugate) elements in L of order 5. From χ(u) = 0 we infer that each 5th root of unity occurs with multiplicity 2 as eigenvalue of t on V (provided 5 | r − 1). Each primitive 11th root of unity occurs with multiplicity 1 as eigenvalue of the elements of order 11 on V (if 11 | r − 1). Assume that there is no regular G-orbit on V . Then r10 ≤ 165(r6 + r4 ) + 220(r4 + 2r3 ) + 396(5r2 ) + 1.440(10r).

120


But this implies that r < 5, a contradiction. All other 9 irreducible characters χi of M11 are treated similarly. (Using that c(M11 ) ≤ 4 only the characters of degree 10, 11, 16 need to be considered.) Let L = M12 , where M(L) and Out(L) have order 2 [Atlas, p. 33]. One checks that c(M12 ) ≤ 5 by considering the maximal subgroups. So as before only the characters χ = χi of small degrees have to be examined, namely χ2 , χ3 of degree 11 and χ4 , χ5 of degree 16 of M12 (which fuse in Aut(M12 )), and the characters of degree 10, 10, 12, 32 of 2.M12 (extendible to 2.M12 .2). In each case there is a regular orbit. The other sporadic groups are treated in the same manner, except when L = J2 is the second Janko group or L = Suz is the Suzuki group. Let us treat these groups (briefly). Let L = J2 . One checks that c(J2 ) ≤ 5. Using Lemma 7.3b this reduces the discussion to the case where E = 2.J2 and V = Vχ with χ = χ22 or χ23 in the Atlas notation [Atlas, p. 43], both characters of degree 6 fusing in 2.J2 .2. Hence X = E is the Schur cover of L and G = G0 = E ◦ Z. The usual counting technique yields the existence of a regular vector when r ≥ 26. Since p > 7 by coprimeness, and since F = Fr contains a square root of 5 by the values of χ on 5-elements, we must have r = p equal to 11 or to 19. On the basis of the table of marks for L = J2 [GAP] we compute (observing that exp (2.J2 )3 = 3) µ2χ,1 = (r − 11)(r + 1)(r + 11)(r2 − 193)/|L|, µ6χ,1 = µ2χ,1 − (r2 − 14r + 265)/1.080, 2 µ10 χ,1 = µχ,1 − 7(r − 11)/300.

This shows that there is a regular orbit for r = 19 but no regular orbit for r = 11. In this latter case the smallest stabilizer H is isomorphic to S3 (as is seen by computing the corresponding polynomials). Since the characters χ22 , χ23 involved are conjugate under a group automorphism of 2.J2 , we have a unique isomorphism type. Let L = Suz. One checks that c(Suz) ≤ 6. As before this reduces the discussion to the case where E = X = 6.Suz and χ = χ115 is of degree 12 [Atlas, p. 130]. Here G = G0 = X ◦ Z and V = Vχ comes from the action of X on the complex Leech lattice. The usual counting argument yields that there is a regular orbit when r ≥ 23. Since p > 13 and 6 | r − 1, it remains the case where r = p = 19. (Since an involution x of 6.Suz belonging to the class 2A0 has the spectral pattern xV = [−18 , 14 ] and |2A0 | = 135.135,


121

the counting method fails for r = 19.) The table of marks for Suz is not available in [GAP], but [Köhler–Pahlings, 2001] indicate, on the basis of a different orbit algorithm, that a regular orbit exists also in the r = 19 case. Remark . As for Theorem 7.1 it suffices to find a (strongly) real vector v such that CG (v) has a regular orbit on V ] (r = 19). The group X = 6.Suz has a subgroup Y ∼ = Z2 × 36 : M11 , which in turn has a subgroup H ∼ = M11 of pure permutations acting transitively on the 12 coordinates of the Leech lattice, a point stabilizer being a maximal L2 (11) subgroup [Atlas, pp. 131, 18]. It follows that ResG H (χ) = 1H +χ5 where χ5 is the irreducible character ∼ of H = M11 of degree 11 (which is rational-valued). Let CV (H) = F v and N = NX (F v). Either H is contained in a 2.M12 subgroup of X, on which χ is irreducible however, or N is in (a conjugate of) Y , with H acting irreducibly on Y /Z(X) ∼ = 35 . We conclude that CX (v) = H and that N = H × Z(X). Application of Lemma 5.1b yields that CG (v) = H. One also easily checks that H has a regular orbit on V . 7.5. Alternating Groups Let L = An for some n ≥ 5, acting on Ω = {1, · · ·, n}. Then Aut(L) = Sn unles n = 6 (Out(A6 ) ∼ = V4 ) and |M(L)| = 2 unless n = 6, 7 (M(A6 ) = M(A7 ) = Z6 ). It is well known that c(x) ≤ n − 1 if x is a transposition (as (12), (13), · · ·, (1n) generate Sn ). For 1 6= x ∈ Sn we generally have (7.5a)

n−2 c(x) ≤ d n−|orb(hxi on Ω)| e + 2,

except when n = 6 and x is an involution of cycle shape (23 ), where c(x) = 5 [Hall et al., 1992]. In addition c(x) ≤ n/2 for n ≥ 7, unless x is a transposition [Guralnick–Saxl, 2003]. As usual we exclude the deleted permutation module, which is the Specht module S (n−1,1) for Sn and which for n ≥ 7 is the unique faithful module of minimal degree. By [Rasala, 1977] for n ≥ 10 the Specht module S (n−2,2) to the partition (n − 2, 2) is the unique faithful module of the second minimal degree, n(n − 3)/2. This module is known to be irreducible for Sn [James–Kerber, 1981, 7.3.23]. Using the branching rule for Specht modules one shows, by induction, that S (n−2,2) is irreducible also for An . From [Schur, 1911, Sec. 50] one knows that for n ≥ 7 the minimal faithful character degrees for 2.An and 2.Sn are 2b(n−2)/2c and 2b(n−1)/2c , respectively.

122


Proposition 7.5b. For n ≥ 8 there exists a regular G-orbit on V . Proof. The character table for An and its related groups is in the [Atlas] for n ≤ 13. Assume first that n ≥ 14. Then by the above (7.5c)

d ≥ n(n − 3)/2

for n ≥ 16, because then 2b

n−2 2 c

> n(n − 3)/2, and d ≥ 64 for n = 14, 15.

Suppose x ∈ Sn is nontrivial with n − |orb(hxi on Ω)| > 3. Then by (7.5a) c(x) ≤ d n−2 4 e + 2 ≤ (n + 9)/4 unless n = 14, x has cycle shape (27 ) and c(x) ≤ 6. If n − |orb(hxi on Ω)| ≤ 3, then x has cycle shape (2), (3), (22 ), (4), (3, 2) or (23 ). Then c(x) ≤ n − 1, n/2, n/2 and,¡in¢ view ¡nof ¢ n (7.5a), three times c(x) ≤ d(n−2)/3e+2 ≤ (n+6)/3. There are , 2· , 2 3 ¡n¢ ¡n¢ ¡n¢ ¡n¢ 3 · 4 , 6 · 4 , 20 · 5 and 15 · 6 elements in Sn of these cycle shapes, respectively. Assume there is no regular G-orbit on V . Let first n ≥ 16. Then by Lemma 7.3b µ ¶ µ ¶ µ ¶ £ n 1 4 n d(1− n−1 n ¤ d(1− 2 ) ) ) d(1− n+9 d n + r + +3 r r /2 ≤ |Sn |r 2 3 4 £ ¡ ¢ ¡ ¢ ¡ ¢¤ 3 + 6 n4 + 20 n5 + 15 n6 rd(1− n+6 ) . Since r > n ≥ 16 and d ≥ n(n − 3)/2 by (7.5c), it follows that 2n! > r4d/(n+9) − r3d/(n+9)+2 − r2d/(n+9)+4 − rd/(n+9)+6 . The second, third and fourth terms on the right are all less than 61 r4d/(n+9) . Consequently nn > 4n! > r4d/(n+9) > n4d/(n+9) and so n > 4d/(n + 9) ≥ 2n(n − 3)/(n + 9). But this implies that n < 12. Let n = 14 or 15. We have the same estimate, replacing the term 4 1 |Sn |rd(1− n+9 ) by |Sn |rd(1− 6 ) . Since then the difference of the terms on the left hand side and the right hand side is an increasing function of d, we may pick d = 64. We get that r64 /2 ≤ 15!r53 + 105r59 + 5.005r56 + 2.761.670r54 . But this implies that r < 14, which is a contradiction (p > 14). Let finally 8 ≤ n ≤ 13. ¡ ¢Recall 1that r ≥ p >2 n. By Lemma 7.3b we may assume that rd ≤ 2 n2 rd(1− n−1 ) + 2n!rd(1− n ) . So for n = 8 only the characters χ = χ3 , χ15 of degrees d = 20, 8, respectively, have to be examined [Atlas, p. 22]. Here, as in all further cases, the counting method works. ¤ It remains to investigate the cases n = 5, 6, 7.


123

n = 5 : The Weil characters of 2.A5 = Sp2 (5) of degree 2 have been already treated in Example 7.3d. Let χ be one of the Weil characters of degree 3 of PSp2 (5) ∼ = A5 . So χ = χ2 or χ3 [Atlas]. Further E = A5 = X and G = G0 = E × Z, and exp (X) = 30. Let V √ = Vχ over F = Fr (so r is any power of a prime p ≥ 7, and F contains 5). As above let r(X) = gcd(r − 1, exp (X)). If r(X) = 2, then r − 1 = 2u for an odd u divisible only by primes larger than 5, and the mark vector on P1 (V ) is the row vector given by (r2 + r + 1, r + 2, 1, 3, 1, 1, 1, 0, 0). This row vector is (r2 + r + 1, r + 2, 3, 3, 1, 1, 1, 0, 0) for r(X) = 6, is (r2 + r + 1, r + 2, 1, 3, 3, 1, 1, 0, 0) for r(X) = 10, and is (r2 + r + 1, r + 2, 3, 3, 3, 1, 1, 0, 0) for r(X) = 30. Using the table of marks for A5 given in Example 7.3d we get: µ2χ,1 = (r − 9)(r − 5)/60, µ2χ,X¯ 2 = (r − 5)/2, µ2χ,X¯ i = 1 for i = 4, 6, 7, µ6χ,1 = (r2 −14r +25)/60, µ6χ,X¯ 2 = (r −5)/2, µ6χ,X¯ i = 1 for i = 3, 4, 6, 7, 10 10 µ10 ¯ 2 = (r − 5)/2, µχ,X ¯ i = 1 for i = 4, 5, 6, 7, χ,1 = (r − 11)(r − 3)/60, µχ,X 30 30 µ30 ¯ 2 = (r − 5)/2, µχ,X ¯ i = 1 for 3 ≤ i ≤ 7. χ,1 = (r − 13)(r − 1)/60, µχ,X

In all other cases µtχ,X¯ i = 0. Using that F13 does not contain a square root of 5, one concludes that there is a regular G-orbit on V ] if and only if r 6= 11. For r = 11 there are point stabilizers of order 2. The character χ = χ4 of A5 [Atlas, p. 2] is afforded by the deleted permutation module, which case we have excluded. (There is no regular G0 -orbit if and only if G0 = S5 × Z with r = 7, in accordance with 5.1a.) The character χ = χ5 of A5 of degree 5, which splits on S5 , is treated by the usual counting argument. For the characters χ = χ8 and χ9 of 2.A5 of degrees 4 and 6, which extend to any 2.A5 .2, the counting argument applies as well. n = 6 : The counting method rules out all cases except the following ones. (i) E = A6 , χ = χ2 resp. χ3 of degree 5, which extend to A6 .21 = S6 but fuse in A6 .22 = PGL2 (9) and A6 .23 = M10 . Here χ2 is the character of A6 afforded by the deleted permutation module. So we have to investigate χ3 only. But χ3 is obtained from χ2 via an (exceptional) outer automorphism of A6 , and the same for the extensions to S6 . So the corresponding reduced pairs are isomorphic. (Therefore they do not appear in the list of Theorem 7.2a !) Of course the polynomials for χ2 , χ3 agree. We know from Example 5.1a that there is no regular G0 -orbit on Vχ , for χ = χ2 or χ3 , if and only if r = 7. From µ6χ,h(12)(34)(56)i = (r − 1)(r − 3)(r − 5)/24 we obtain two orbits with point stabilizers of order 2.

124


(ii) E = 2.A6 , χ = χ8 resp. χ9 of degree 4, which extend to both isoclinic variants X = 2.A6 .21 and fuse in the 2.A6 .22 . This has been already treated in Example 5.7b, but let us argue also on the basis of Proposition 7.3c. For χ = χ9 and X = 2+ S6 the polynomials have 7 as the only positive root coprime to |E|, and µ6χ,1 = (r − 3)(r − 7)(r + 11)/720. Hence there is no regular orbit for r = 7. Considering the subgroups ¯ 3 = h(123)(345)i, X ¯ 4 = h(123)(45)i and X ¯ 4 = h(123)(456), (123)(45)i X ¯ of X = S6 we have 6 µ6χ,X¯ 3 = (r − 1)/6, µ6χ,X¯ 4 = µχ, ¯ 5 = 1. X

Hence there are just three G-orbits on V ] for r = 7, and the point stabilizers are isomorphic to Z3 , Z6 , Z3 × S3 . This is as stated in Theorem 7.2a, and known from Example 5.7b. We have also seen that there exists no real vector in this case. (iii) There are four algebraically conjugate characters, χ, of E = 3.A6 of degree 3 which pairwise fuse in 3.A6 .21 , 3.A6 .22 or 3.A6 .23 . So we have to consider X = E and G = G0 = X ◦ Z. The splitting fields F = Fr for χ require the 3rd roots of unity and the square roots of 5. The counting argument yields that there are regular orbits when r ≥ 53. Hence if there are no regular vectors, then r = 19, 31 or 49. It suffices to compute: µ6χ,1 = (r − 19)(r − 25)/360, µ6χ,h(12)(34)i = (r − 15)/4; 30 µ30 χ,1 = (r − 13)(r − 31)/360, µχ,h(12)(34)i = (r − 27)/4.

It follows that there is no regular orbit for r = 19 and r = 31, in which cases there are orbits with stabilizer of order 2. For r = 49 observe that 6 ] there are r−1 6 · µχ,1 (47) = 16 regular E ◦ Z6 -orbits on V by Proposition 7.3c. Using that exp (E) = 22 ·3·5 and that each involution of G is in E ◦Z6 we see that these orbits fuse to 2 regular orbits for G. So the polynomial 6 µ48 χ,1 = µχ,1 depends only on t = 6. (iv) For E = 3.A6 there is a pair of algebraically conjugate irreducible characters χ = χ15 of degree 6 [Atlas, p. 5], which extend to X = 3.A6 .23 (requiring a square root of −2). So G = G0 = X ◦ Z. Assuming that there is no regular orbit, 1 r6 ≤ |2A0 |(r4 + r2 ) + |3A0 B0 |(3r2 ) + |5A0 B0 |(r2 + 4r) 2 = 45(r4 + r2 ) + 120r2 + 36(r2 + 4r).


125

This implies that r < 8, hence r = 7. This is ruled out by the method of Proposition 7.3c; in fact, the polynomial µ6χ,1 has only the root 1. n = 7 : The counting argument rules out all cases except when E = 2.A7 , χ = χ10 resp. χ11 of degree 4, which fuse in 2.S7 . We get regular orbits for r ≥ 23. Since p > 7 and a splitting field F = Fr for χ requires a square root of −7, and since SL4 (17) and SL4 (19) are 70 -groups, we must have 2 r = 11. In this case the multiplicities µ10 χ,1 = (r − 11)(r + 12r − 7)/2.520 10 and µχ,h(123)(456)i = (r − 5)/6 give the result as stated in Theorem 7.2a.

7.6. Linear Groups Let L = Lm (q) = PSLm (q) for some integer m ≥ 2 and some prime power q. Since L2 (3) is solvable and L2 (4) = L2 (5) ∼ = A5 , L2 (9) ∼ = A6 and L4 (2) ∼ = A8 , we may and do exclude the pairs (m, q) = (2, 3), (2, 4), (2, 5), (2, 9) and (4, 2). [Guralnick and Saxl, 2003] have given upper bounds for c(L). Ignoring the groups excluded one has c(Lm (q)) ≤ 4, 4, 6 for m = 2, 3, 4, respectively, and (7.6a)

c(Lm (q)) ≤ m

in all other cases. From Appendix (C1) we know that ½ (7.6b)

R0 (Lm (q)) =

(q − 1)/gcd(q − 1, 2) if m = 2 (q m − 1)/(q − 1) if m > 2

except when (m, q) = (3, 2), (3, 4), (4, 3) (again ignoring the groups already ruled out). One has R0 (L3 (2)) = 3, R0 (L3 (4)) = 6 and R0 (L4 (3)) = 26, and otherwise R0 (Lm (q)) is just attained by the degree of Weil characters. ¯ ⊆ Aut(L). The Schur multiplier and the automorphism Recall that G group of L (and of any group of Lie type) has been determined in [Steinberg, 1967]. Each automorphism is a product of an inner, a diagonal, a field and a graph automorphism. In the present case the graph automorphism (of the Dynkin diagram) is just the inverse transpose automorphism, and the diagonal automorphisms are induced by conjugation with the diagonal matrices in GLm (q) with determinant 6= 1. So Out(L) and M(L) are known.

126


Proposition 7.6c. If there is no regular G-orbit on V , then m ≤ 3 or L∼ = L4 (3) and d ≤ 52, L ∼ = L5 (2) and d = 30, or L ∼ = L6 (2) and d = 62. This readily follows from the above information by applying Lemma 7.3b. Note that L6 (2) is not an Atlas group. We know from Theorem 4.5b that L6 (2) has a Weil character of degree d = 26 − 2 = 62. This is the unique irreducible character of this degree; the second lowest degree is 217. m = 2 : Suppose that there is no regular G-orbit on V . Let first q be even. Then Lemma 7.3b implies that q ≤ 16, hence q = 16 or q = 8. From the character table one reads off that, for each character χ 6= 1L , of L = L2 (16) or a possible extension to Aut(L) = L.4, no element of prime order in Aut(L) has eigenspaces of dimension greater than χ(1) − 6. Thus (5.6d) yields that r6 ≤ 2|Aut(L)| in this case, giving the contradiction r < 6. It remains to exclude the case L = L2 (8). Consider for instance the character χ = χ2 of degree d = 7 of E = L2 (8) extendible to X = Aut(L) = L.2 [Atlas, p. 6], letting G = X × Z and V = Vχ over F = Fr . Determining spectral patterns of elements of prime order we get 1 1 1 r7 ≤ |2A|(r4 + r3 ) + |3A|(2r3 + r) + |7ABC|(7r) + |3B|(r3 + 2r2 ). 2 2 2 Here |2A| = 63, |3A| = 56, |7A| = |7B| = |7C| = 72, |3B| = 84, which gives r7 ≤ 63r4 + 161r3 + 84r2 + 784r. We obtain that r < 5, which is a contradiction since |L| = 23 · 32 · 7. The other characters are treated similarly. So let q be odd, q = sf for some (odd) prime s. Assume q ≥ 37. Consider first the case where f = 1 (q = s). We assert that then d ≥ q − 1. For otherwise χ is one of the Weil characters ξi 6= ξ¯i of Sp2 (q) for i = 1, 2, of degree (q ± 1)/2, where the pairs of algebraically conjugate characters must fuse in GL2 (q) (see Theorem 4.5a and Appendix C2; the character ¯∼ table of Sp2 (q) can be found in [Schur, 1911]). Hence G = L = L2 (q). From √ Theorems 4.4 and 4.5a we infer that χ takes the value (−1 ± ±q)/2 on the noncentral elements of order q. So these elements have (q ± 1)/2 distinct eigenvalues on V = Vχ , whence no eigenspace has dimension greater than 1. For a noncentral element g ∈ Sp2 (q) of prime order 6= q we have χ(g) = 0 or ±1, which implies that no eigenspace has dimension greater than (q + 5)/4. Since q = s and |L| = q(q 2 − 1)/2, it follows that r(q−1)/2 ≤ q(q 2 − 1)/2 · q (q+5)/4 .


127

From q ≥ 37 we infer that r < 5, a contradiction. So we indeed have ¡ ¢ 4 d ≥ q − 1. Using c(L) ≤ 4, Lemma 7.3b yields that r ≤ q(q 2 − 1) q−1 , which again leads to a contradiction. Hence we must have f > 1. Using that |Aut(L)| = f q(q 2 − 1) from Lemma 7.3b we now get ¡ ¢8/(q−1) r ≤ 2f q(q 2 − 1) , and this gives a contradiction as before. Consequently q < 37, that is, q ≤ 31. Now L = L2 (q) is an Atlas group, and with the usual counting arguments one can rule out all cases except for q = 7. So let L = L2 (7). Let χ be one of the Weil characters χ2 , χ3 = χ ¯2 of L of degree 3, which fuse in PGL2 (7) [Atlas, p. 3]. Hence G = G0 = L × Z in this case. If there is no regular G-orbit on V = Vχ , then 1 1 r3 ≤ |2A|(r2 + r) + |3A|(3r) + |7AB|(3r) = 21r2 + 129r. 2 3 Thus r < 26. Since χ, and hence F = Fr , requires a square root of −7, this implies that r = 11, 23 or 25. Let H be generated by an involution of L (in the unique class 2A). One finds: µ2χ,1 = (r − 9)(r − 11)/168, µ2χ,H = (r − 7)/4, µ6χ,1 = (r2 − 20r + 43)/168, µ6χ,H = (r − 7)/4. Hence there is no regular orbit when r = 11, in which case there are r−1 2 · (r − 7)/4 = 6 orbits with point stabilizers of order 2 (conjugate to H). Now gcd(r − 1, exp (L)) is 2 for r = 23, and is 12 for r = 25. It follows that there are regular orbits in these cases. For r = 25 observe that the r−1 6 6 · µχ,1 (r) = 4 regular orbits for L × Z6 must fuse in G = L × Z24 since each involution in G is contained in L × Z6 . Let next χ = χ6 be of degree 6, which extends to X = PGL2 (7) (affording a square root of 2). We have to consider G = X × Z, and the counting method yields that there is a regular orbit when r ≥ 11. Hence the only questionable case is r = 5, which is ruled out by showing that µ4χ,1 > 0 at r = 5. For all other characters the counting method applies and yields that there are regular orbits. m = 3 : Using that c(L) ≤ 4 and d ≥ R0 (L) = (q 3 − 1)/(q − 1), application of Lemma 7.3b shows that a regular orbit exists unless q ≤ 5. Thus we only have to examine the cases L = L3 (3) and L = L3 (4).

128


Let L = L3 (3). If d 6= 12 or 13, no prime order element of Aut(L) has an eigenspace of dimension greater than d − 8 [Atlas, p. 13]. Then r8 ≤ 2|Aut(L)| forces that r < 4. Hence there is a regular orbit. For the characters χ = χ2 and χ3 of degrees 12 and 13 the usual counting method works. Let L = L3 (4). Recall that R0 (L) = 6, and this minimal degree just happens for the faithful irreducible characters χ = χ41 of X =√6.L.21 [Atlas, √ p. 25]. This χ is rational-valued on E = 6.L and requires 2 or −2 on X (depending on the isoclinism type of X). Of course also 6 | r − 1. Let us describe the situation carefully. We have exp (X) = 23 · 3 · 5 · 7. The conjugacy class 2A of L lifts to two conjugacy classes 2A0 , 2A3 of involutions in E (χ(2A0 ) = −2, χ(2A3 ) = 2). Since Z(E) ⊆ Z and |2A| = 315, this gives the contribution 315(r4 + r2 ). From χ(2B0 ) = 0 and |2B| = 280 we get the contribution 280(2r3 ). From χ(3A0 ) = 0 and |3A| = 2.240 we get the contribution 1.120(3r2 ). The two conjugacy classes 5AB of L (|5A| = |5B| = 4.032) lift to two conjugacy classes 5A0 B0 of elements of order 5 in E, with χ(5A0 ) = χ(5B0 ) = 1. Hence the contribution 1.120(r2 + 4r) if 5 | r − 1 and 1.120(r2 ) otherwise. The two conjugacy classes 7AB of L lift to two conjugacy classes 7A0 B0 of order 7, with χ(7A0 ) = χ(7B0 ) = −1. From |7A| = |7B| = 2.280 we obtain the contribution 960(6r) if 7 | r − 1 and 960r otherwise. Thus if there is no regular orbit, then r6 ≤ 315r4 + 560r3 + 4.795r2 + 10.240r. This forces that r < 19. Since 3 | r −1 and r > 7, we must have r = 13. But ±2 are not squares mod 13. Therefore G = E ◦ Z or G = G0 = (E ◦ Z).2 (and r = 13). The table of marks for L = L3 (4) is contained in the [GAP] library, and one gets that the polynomial µ12 χ,1 does not vanish at r = 13. So there is a regular vector v ∈ V for E ◦ Z, and this is regular also for G0 since otherwise G0 = (CG0 (v)E) ◦ Z. All other irreducible characters for groups n.L3 (4).a can be treated by the usual counting method. m ≥ 4 : By the proposition only the following cases have to be treated. (i) L = L4 (3) and d ≤ 52: In all cases where 26 < d ≤ 52, no eigenspace of a noncentral prime order element of G has dimension greater than d − 13 [Atlas]. Hence r13 ≤ 2 · |L| · 8 forces that r < 5. For the characters χ = χ2 resp. χ3 of L, which fuse in L.21 but split on X = L.22 , we similarly


129

conclude from r26 ≤ |2D|r20 + |L.22 |2r16 that necessarily r < 7. Thus there is a regular orbit for G = X ◦ Z. (ii) L = L5 (2) and d = 30: Here χ = χ2 is the unique minimal (Weil) character (of degree 25 − 2) of L discussed in Theorem 4.5b. It is rational and extends to X = L.2 requiring a square root of 2 [Atlas, p. 70]. So G = L × Z or G = X × Z. The largest dimension of an eigenspace on V of a prime order element in X is 22 [Atlas]. From r8 ≤ 2 · |X| we obtain that r < 9. There are regular orbits. (iii) L = L6 (2) and d = 62: Then again χ = ξ is the Weil character of L studied in Theorem 4.5b (see also Appendix C1). Since L6 (2) is not an Atlas group, we must go into some details. By uniqueness ξ extends to X = L.2 = Aut(L), and we consider G = G0 = X × Z. We assert that dim F CV (g) ≤ 48 for each noncentral element g ∈ G of prime order. In order to prove this we may enlarge F , if necessary, by adjoining the 8th roots of unity. (If F0 is an extension field of F and V0 = F0 ⊗F V , then dim F0 CV0 (g) = dim F CV (g).) Embed L into a standard holomorph T to Q = 21+12 . By (4.5b) there is a subgroup hL, τ i of T mapping onto X and + satisfying τ 2 ∈ Z(Q). There is z ∈ Z such that (zτ )2 = 1, so that X and G appear as subgroups of T Z. Since ±2 are squares in F , by Theorem 4.3e there is a faithful irreducible F [T Z]-module W (of F -dimension 26 ), and V is a constituent of its restriction to G since W affords the character ξ + 2 · 1L on L by Theorem 4.5b. Applying Theorem 6.2c we get dim F CV (g) ≤ dim F CW (g) ≤

3 dim F W = 48, 4

as asserted. There is a regular since r62 > 2 · |L.2|r48 for r ≥ 11. 7.7. Symplectic Groups Let L = S2m (q) = PSp2m (q) for some integer m and some prime power q. Since S2 (q) = L2 (q) and S4 (2) = A6 are already treated, we assume that m ≥ 2 and exclude the pair (m, q) = (2, 2). From [Guralnick–Saxl, 2003] one knows that then c(L) ≤ 2m + 1 (and c(L) ≤ 2m if q is odd and m ≥ 3). Up to few exceptions |M(L)| = (2, q − 1), and this is also the order of the group of outer diagonal automorphisms, and for m ≥ 3 there is no proper

130


graph automorphism [Steinberg, 1967]. As stated in Appendix (C2), one has d ≥ (q m −1)/2 if q is odd and d ≥ (q m −1)(q m −q)/(2(q +1)) otherwise. Proposition 7.7a. There is a regular G-orbit on V except possibly when m ≤ 6 and one of the following holds: • (m, q) = (2, 3) and d ≤ 30, (m, q) = (2, 4) and d ≤ 34, (m, q) = (2, 5) and d ≤ 40, (m, q) = (3, 2) and d ≤ 35, or (m, q) = (4, 2) and d ≤ 85. • (m, q) = (2, 7), (2, 9), (2, 11), (3, 3), (3, 5), (4, 3), (5, 3), (6, 3) and χ is one of the irreducible Weil characters of the symplectic group. This follows from the above information by applying Lemma 7.3b. The groups S4 (7), S4 (9), S4 (11) and S6 (5), S8 (3), S10 (3), S12 (3) are not Atlas groups. From Theorem 4.5a we know a great deal about Weil characters of symplectic groups (see also Appendix C). We need some more details on fixed point ratios. Let E = S2m (q) or E = Sp2m (q) for some m ≥ 2 and some odd q = q0f (q0 prime), assuming that χ is on E one of the irreducible Weil characters ξ1 , ξ2 of degree (q m − 1)/2 and (q m + 1)/2, respectively. Recall that just one of these characters is faithful for Sp2m (q), and that ξ1 is faithful if and only if q m ≡ 1 (mod 4). The characters are not invariant in E.2 and require a square root of ±q if q is not a square Theorem 4.5a. In particular G = G0 = E ◦ Z. Let U be the standard module for S = Sp2m (q). Lemma 7.7b. Suppose the character χ of G afforded by V is a Weil character on the (symplectic) core E. Let g ∈ G be a noncentral element of prime order, say s. 1 (q m + (s − 1)q m−1 ) if s 6= q, and (i) If s is odd, then dim F CV (g) ≤ 2s m−1 m−1 otherwise dim F CV (g) ≤ q − 1 or q depending on whether χ = ξ1 or χ = ξ2 on E.

(ii) Suppose E = L = S2m (q) and that g is an involution in L lifting to an element of order 4 in S = Sp2m (q). Then χ(g) = ±1. (iii) Suppose either that g = g 0 is a (noncentral) involution in E = S = Sp2m (q) or that E = S2m (q) and g is the image of such an involution g 0 in 0 S. Let |CU (g 0 )| = q 2m , so that m0 is an integer with 1 ≤ m0 < m, and let m00 = m − m0 . Then CS (g 0 ) ∼ = Sp2m0 (q) × Sp2m00 (q), and 0 00 0 00 1 1 ξ1 (g 0 ) = ± (q m − q m ) , ξ2 (g 0 ) = ± (q m + q m ), 2 2 00

the positive signs holding precisely when q m ≡ 1 (mod 4).


131

Proof. Pick ξ1 , ξ2 such that ξ = ξ1 + ξ2 is a Weil character of S (and χ = ξ1 or ξ2 on E). Let j be the central involution of S, as usual. Let T be the standard holomorph to Q = (q0 )1+2mf (Theorem 4.3c). Recall + from Theorem 4.5a that Q : S is a subgroup of T , that U = Q/Z(Q) is the standard module for S, and that ξ is the restriction to S of some faithful irreducible character of T , which will be also written ξ. Note that q is a divisor of |E| and so q0 6= p by coprimeness. Enlarging F if necessary (so that q0 | r − 1) there is an F T -module W affording ξ. Write ResTS (W ) = V1 ⊕ V2 with Vi affording ξi . Embed Z(Q) = Z(T ) into Z such that ξ and χ lie over the same linear character of Z(Q). (i) Since s is odd, we may identify g = xz where x ∈ S and z ∈ Z is an sth root of unity. Let z act trivially on T , and extend ξ to a faithful character of Tz = T hzi such that ξ(z)/ξ(1) = χ(z)/χ(1). (We have Tz = T if s = q0 and Tz = T × hzi otherwise.) From Theorem 6.2c it follows that dim F CW (g) ≤ b 1s (q m + (s − 1)q m−1 )c if s 6= q and dim F CW (g) ≤ 2q m−1 otherwise. Clearly CW (g) = CV1 (g) ⊕ CV2 (g). By Theorem 4.5a, ξ2 = ξ1 +1S on 20 -elements of S. The assertion follows by using that dim F CW (g) = hξ, 1ihgi = hξ, νihxi where ν is the linear character of hxi defined by ν¯(x) = χ(z)/χ(1), and similar statement for the Vi , ξi . Note also that b(2q m−1 − 1)/2c = q m−1 − 1 and b(2q m−1 + 1)/2c = q m−1 . (ii) Let ge ∈ S have image g, so that ge2 = j. Let V = Va afford the Weil character ξa which is not faithful for S, and let Vb afford the the other irreducible Weil character ξb (with ξ = ξa + ξb ). Then j acts as −1 on Vb , so that CVb (e g ) = 0 and CV (g) = CW (e g ). By Theorem 4.5a, ξ(e g ) is an integer, and ξa (g) = ξa (e g ) is an integer as g is an involution. Hence ξb (e g ) is an integer. Since ge2 = j, the eigenvalues of ge on Vb are ±i. We infer that ξb (e g ) = 0. Further |CU (e g )| = 1 and so ξ(e g ) = ±1 by Theorem 4.4, because ge is good for U . Hence χ(g) = ξa (g) = ξ(e g ) = ±1. (iii) Notice that g 0 and g 0 j belong to different conjugacy classes of S (but both mapping onto g if g 6= g 0 ). So the following depends on the choice of the inverse image. We have a proper (orthogonal) decomposition U = [U, g 0 ] ⊕ CU (g 0 ) = U − ⊕ U + , where U − , U + are the eigenspaces of g 0 to the eigenvalues −1 and +1, respectively. This corresponds to a central 0 00 f f decomposition Q = Q0 ◦ Q00 , where Q0 = (q0 )1+2m and Q00 = (q0 )1+2m , + + and yields embeddings of the standard holomorphs, and of the “Young group” Y = Sp2m0 (q) × Sp2m00 (q) into S. Since ξ is (absolutely) irreducible

132


on (Q0 : Sp2m0 (q)) ◦ (Q00 : Sp2m00 (q)), we obtain a tensor decomposition ResY (ξ) = ξ 0 ⊗ ξ 00 of Weil characters. Just by considering character degrees we get that ResY (ξ1 ) = ξ10 ⊗ ξ200 + ξ20 ⊗ ξ100 and ResY (ξ2 ) = ξ10 ⊗ ξ100 + ξ20 ⊗ ξ200 . Here the irreducible Weil characters have to be chosen properly. One knows that there are precisely m − 1 conjugacy classes of noncentral involutions in S = Sp2m (q), the conjugacy class of an involution being determined by the order of its fixed point group on U [Dieudonné, 1971, pp. 25, 26]. This readily gives the statement for CS (g 0 ). Moreover, the conjugacy class of g 0 is determined by the integer m0 , and g 0 is conjugate in S to the element 10 × j 00 in Sp2m0 (q) × Sp2m00 (q), j 00 being the central involution in Sp2m00 (q). The character values are obtained by using that ξ100 00 is faithful on Sp2m00 (q) by Theorem 5.4a, satisfying ξ100 (j 00 ) = −(q m − 1)/2, 00 if and only if q m ≡ 1 (mod 4). ¤ We now discuss the cases remaining by virtue of Proposition 7.7a. m = 2 : We have to examine the following. (i) L = S4 (3) and d ≤ 30: Let us consider first the Weil characters. So let χ be faithful for E = Sp4 (3) of degree 4 (characters χ21 , χ22 in [Atlas, p. 27]). We have G = G0 = E ◦ Z, and F = Fr requires a square root of −3, that is, r − 1 is divisible by 3. Of course r is coprime to |L| and so not divisible by 2, 3, 5. The usual counting argument (on the basis of the character table) yields that there is a regular G-orbit on V if r ≥ 61. Hence r ∈ {7, 13, 19, 31, 37, 43, 49}. We proceed by computing the relevant polynomials µtχ,1 . The table of marks for L ∼ = U4 (2) is in the [GAP] library. We obtain: µ6χ,1 = (r − 7)(r − 13)(r − 19)/|L|, µ12 χ,1 = (r + 11)(r − 13)(r − 37)/|L|, 2 µ30 χ,1 = (r − 31)(r − 8r + 223)/|L|.

We conclude that there is no regular orbit if and only if r ∈ {7, 13, 19, 31, 37}. For r = 43 note that gcd(r − 1, exp (E)) = 6, and for r = 49 by Proposition 7.3c we have 4 · µ12 χ,1 (49) = 4 regular orbits for E ◦ Z12 which fuse in G = E ◦ Z48 as each involution of G is in E ◦ Z12 . ¯ = L leads Computing the polynomials for nontrivial subgroups of X to the results stated in Theorem 7.2a. We know from Corollary 7.2b that


133

for r = 31 the pair (G, V ) admits a real (but not strongly real) vector. The information given in Sec. 7.1 for r = 7, 13, 19, basically obtained by a study of the [Atlas], manifest that we have six nonreal reduced pairs here. Let next χ = χ2 or χ3 for E = L = S4 (3), a Weil character of degree d = 5. Here G = G0 = L × Z and 3 | r − 1. By the counting principle we get regular orbits unless r < 56. So r is as before. Computation with [GAP] yields µ2χ,1 = (r − 7)(r − 9)(r − 13)(r − 15)/|L|, µ6χ,1 = (r − 5)(r − 7)(r − 13)(r − 19)/|L|, 2 µ10 χ,1 = µχ,1 −

1 5

1 6 and µ30 χ,1 = µχ,1 − 5 .

We conclude there are no regular orbits if and only if r ∈ {7, 13, 19}. We similarly get the minimal stabilizers as stated in Theorem 7.2a. Let χ = χ4 in the Atlas notation, a rational (Weil) character of L ∼ = U4 (2) of degree 6 which extends rationally to X = L.2. So G = G0 = X ×Z. The counting argument yields the existence of a regular orbit unless r < 47. The method of Proposition 7.3c shows that there are no regular orbits if and only if r ∈ {7, 11, 13}, and one gets the minimal stabilizers listed in Theorem 7.2a. For the remaining projective representations of S4 (3) the counting principle works. (ii) L = S4 (4) and d ≤ 34: Consider first the minimal character χ = χ2 of degree 18, which is rational and extends to X = Aut(L) = L.4 requiring √ −1 [Atlas, p. 45]. Assuming that there is no regular orbit we have r18 ≤ |2AB|(r12 + r6 ) + |2C|(r10 + r8 ) + |2D|(2r9 ) + |L|(3r6 ) = 510(r12 + r6 ) + 3.825(r10 + r8 ) + 2.720r9 + 28 · 33 · 52 · 7r6 . This implies that r < 9, which is impossible (by coprimeness). For the representations of L.2 of dimension 34, no prime order element has an eigenspace of dimension greater than 22, and the result follows. (iii) L = S4 (5) and d ≤ 40: For the minimal Weil characters (of degree 12) of 2.L = Sp4 (5) the counting argument (based on the character table [Atlas, pp. 62–63]) gives a regular orbit (unless r < 6). For the other Weil characters (of degree 13) of L the result is similar. For the character χ = χ4 of L of degree 40, which extends to L.2, the largest dimension of an eigenspace of a prime order element is 26, and the result follows.

134


(iv) L = S4 (7) and χ is a Weil character: Here L is not an Atlas group. We argue on the basis of Lemma 7.7b. By Theorem 4.5a, F requires a square root of −7. Either χ = ξ1 is faithful for E = Sp4 (7) (of degree d = (72 − 1)/2 = 24) and G = G0 = E ◦ Z, or χ = ξ2 , d = 25 and G = G0 = L × Z. Let g ∈ G be a noncentral element of prime order s. 1 (72 + (s − 1)7) if s 6= 7 If g is not an involution, then dim F CV (g) ≤ 2s and dim F CV (g) ≤ 7 otherwise. So the fixed point ratio f (g, V ) < 21 , at any rate. Suppose next that χ = ξ2 and x is an involution in E = L that lifts to an element of order 4 in Sp4 (7). Then χ(x) = ±1 by the lemma, 13 when g = xz for some z ∈ Z. Let finally x ∈ E be a whence f (g, V ) ≤ 25 noncentral involution which is not of this type. By the lemma

1 χ(x) = − (7 + 7) = −7 2 if χ = ξ2 , and χ(x) = 0 otherwise. So f (g, V ) ≤ 21 when χ = ξ1 (for any g), in which case there is a regular orbit since r12 > 2|Sp4 (7)| for r ≥ 6. So 16 for each g, so that we have to inspect when let χ = ξ2 . Then f (g, V ) ≤ 25 the inequality r9 > 2|S4 (7)| = 29 · 32 · 52 · 74 does hold. This indeed holds for r ≥ 9, implying that there is a regular orbit also in this case. (v), (vi) L = S4 (9) and L = S4 (11), with χ being a Weil character: These are treated in exactly the same manner as (iv). m = 3 : By Proposition 7.7a the following cases have to be examined. (i) L = S6 (2) and d ≤ 35. Consider first the character χ = χ2 of L of degree 7 [Atlas, p. 47]. Here we have G = G0 = L × Z. The counting argument shows that there is a regular orbit when r ≥ 80. So r = p can be any prime between 11 and 79 at this stage (by coprimeness). We use projective marks. The table of marks for L = Sp6 (2) is in [GAP]. We get µ2χ,1 = (r − 5)(r − 7)(r − 9)(r − 11)(r − 13)(r − 17)/|L|, µ6χ,1 = (r − 7)(r − 13)(r − 19)(r3 − 23r2 + 187r − 325)/|L|, 2 µ14 χ,1 = µχ,1 −

1 7

1 6 and µ42 χ,1 = µχ,1 − 7 .

We deduce that there are no regular orbits if and only if r ∈ {11, 13, 17, 19}. The minimal stabilizers listed in Theorem 7.2a are obtained from the corresponding polynomials to the appropriate nontrivial subgroups of L.


135

Consider next the character χ = χ31 of degree 8 of E = 2.Sp6 (2). The counting method yields that there are regular orbits for G = E ◦ Z except possibly when r = 11. This is handled by showing that the polynomial µ10 χ,1 is positive at r = 11. (ii) L = S6 (3) and χ is a Weil character: Here L is an Atlas group [Atlas, pp. 110–113), and the usual counting arguments work. There is a regular orbit. (iii) L = S6 (5) and χ is a Weil character: Here L is not an Atlas group. By Theorem 4.5a the field F requires a square root of 5. Either χ = ξ1 is faithful (of degree (53 − 1)/2 = 62) for E = Sp6 (5) and G = G0 = E ◦ Z, or χ = ξ2 and G = G0 = L × Z. If g is a noncentral element in G of odd prime order, then dim F CV (g) ≤ 61 (53 + 2 · 52 ) by Lemma 7.7b (taking the worst case o(g) = 3), whence f (g, V ) < 21 . If G = L×Z and g ∈ L is an involution which lifts to an element of order 4 in Sp6 (5), we have χ(g) = ξ2 (g) = ±1 32 for any z ∈ Z. If g ∈ L is an involution by the lemma and so f (gz, V ) ≤ 63 which is not of this type, then χ(x) = ξ2 (x) = 15. If χ = ξ1 and x ∈ E is a noncentral involution, then χ(g) ∈ {10, −10} by Lemma 7.7b. 36 for any noncentral element g ∈ G if We conclude that f (g, V ) ≤ 62 39 G = E ◦ Z, and f (g, V ) ≤ 63 if G = L × Z. In the second (worse) case we verify that r24 > 2|S6 (5)| = 210 · 34 · 59 · 7 · 13 · 31

for r ≥ 5. Hence there are regular orbits. m = 4 : By Proposition 7.7a the following has to be treated. (i) L = S8 (2) and d ≤ 85: Then d = 35, 51 or 85 [Atlas, p. 124]. If d = 35 or 85 (characters χ3 , χ4 in the [Atlas]), then no prime order element has an eigenspace on V of dimension greater than d − 15. From r15 ≤ 2 · |S8 (2)| we obtain that r < 6. Hence there is a regular orbit. For d = 35 (character χ2 ) no prime order element, except the involutions in class 2A, have eigenspaces of dimension greater than 23, and the involutions in 2A have the spectral pattern [−1(28) , 1(7) ]. From r35 ≤ |2A|r28 + 2|L|r23 = 255r28 + 217 · 35 · 52 · 7 · 17 we obtain that r < 9. Hence there is again a regular orbit. (ii) L = S8 (3) and χ is a Weil character: The field F requires a square root of −3 by Theorem 4.5a, and ξ1 is faithful for E = Sp8 (3) (of degree

136


40). So G = G0 = E ◦ Z if χ = ξ1 , and G = G0 = L × Z if χ = ξ2 . Let g be a noncentral element of G of prime order s. If s is odd, by Lemma 7.7b we have dim F CV (g) ≤ 33 − 1 = 26 when χ = ξ1 and dim F CV (g) ≤ 27 otherwise, considering the worst case s = 3. If χ = ξ2 and g ∈ L is an involution lifting to an element of order 4 in Sp8 (3), then χ(g) ± 1 by the lemma. If g ∈ L is an involution which is not of this kind, then χ(g) = ξ2 (g) ∈ {9, −15}. If χ = ξ1 on E and g ∈ E is a (noncentral) involution, then χ(g) ∈ {0, 12, −12} by Lemma 7.7b. 26 when χ = ξ1 , and f (g, V ) ≤ We conclude that, for all g, f (g, V ) ≤ 40 when χ = ξ2 . Assuming that there is no regular G-orbit on V , in the first case we have 28 41

r15 ≤ 2|Sp8 (3)| = 216 · 316 · 52 · 7 · 13 · 41, which forces that r < 16. Since r is coprime to 13 (and to 2 · 3 · 5 · 7) and since −3 is not a square modulo 11, we obtain the desired contradiction. In the second case the inequality r13 ≤ 2|S8 (3)| only holds if r < 21, and it remains to rule out the case r = 19. We have to improve the estimate. By (5.6d) we may replace the factor 2 by (1 + r115 ), which is not sufficient, however. Consider the conjugacy class 2B of involutions in S8 (3) where χ = ξ2 takes the value −15. By Lemma 7.7b, |2B| = |Sp8 (3)|/(|Sp6 (3)|·|Sp2 (3)|) = 2 · 36 · 5 · 41. We may replace the above inequality by r41 ≤ |2B|(r28 + r13 ) + |S8 (3)|(r27 + r14 ). This implies that r < 19, as desired. m = 5, 6 : By Proposition 7.7a we have merely to treat the Weil characters for Sp2m (3). One argues as for Sp8 (3). 7.8. Unitary Groups Let L = Um (q) = PSUm (q) for some integer m ≥ 3 and some prime power q. The case (m, q) = (3, 2) cannnot happen since L is simple, and we exclude (m, q) = (4, 2) since U4 (2) ∼ = S4 (3) has been already treated in Sec. 7.7. The automorphism group of L, and its Schur multiplier, is known [Steinberg, 1967]. Recall that the centre Z(SUm (q)) is (cyclic) of order


137

gcd(q + 1, m). By [Guralnick–Saxl, 2003] c(L) ≤ m + 1 except when m = 4, in which case c(L) ≤ 6, and c(L) ≤ m for m ≥ 5. By [Tiep–Zalesskii, 1996] ½ R0 (Um (q))) =

(q m − q)/(q + 1) if m is odd , (q m − 1)/q + 1) otherwise

except when (m, q) = (4, 3) (see Appendix C3). Then if d = χ(1) ≤ R0 (L) + 1, χ is a Weil character of SUm (q). Using that d ≥ R0 (L) one gets the following, on the basis of Lemma 7.3b. Proposition 7.8a. There is a regular G-orbit on V except possibly in the following cases: • (m, q) = (3, 3) and d ≤ 21; (m, q) = (4, 3) and d ≤ 45; (m, q) = (5, 2) and d ≤ 55; (m, q) = (6, 2) and d ≤ 56. • (m, q) = (3, 4), (3, 5), (3, 7), (4, 4), (5, 3), (7, 2), (8, 2), (9, 2) and χ is one of the Weil characters of SUm (q). The groups U4 (4), U5 (3), U7 (2), U8 (2) and U9 (2) are not Atlas groups. So we need some general information on their Weil characters. Some crucial facts have been described in Appendix (C3). Let U be the standard module for Gu = GUm (q), and let λu be a fixed linear character of Zu = Z(Gu ) of order q + 1 (noting that Zu is cyclic of order q + 1). There is a generic Weil character ξu of Gu , which is rational-valued and satisfies ξu2 = πU . There are precisely q + 1 irreducible constituents ξ0 , ξ1 , · · ·, ξq of ξu , all remaining irreducible and distinct when restricted to Su = SUm (q). The Weil character ξj is the (irreducible) constituent of ξu lying above the linear Pq q m +(−1)m q and character λju , so that ξu = j=0 ξj . We have ξ0 (1) = q+1 m 2πi/(q+1) ξj (1) = ξ0 (1) − (−1) for j > 0. Let ε = e , and let zu be the generator of Zu satisfying λu (zu ) = ε. Lemma 7.8b. Suppose the character χ of G afforded by V agrees on the (unitary) core E with an (irreducible) Weil character. Let g ∈ G be a noncentral element of prime order s. (i) Suppose g induces on E an inner or diagonal automorphism, and 1 (q m + (s − 1)q m−1 ) when assume that s - q + 1. Then dim F CV (g) ≤ s(q+1) 2 q m otherwise. s - q or s = 2 | q, and dim F CV (g) ≤ q+1 (ii) Suppose g is an involution inducing an outer field automorphism on E. Then either χ(g) = 0 or χ(g) = ±q a for some integer a ≤ m 2.

138


Proof. Let q = q0f , q0 prime. Since q is a divisor of |E|, q0 6= p by coprimeness. Enlarging F , if necessary, we may assume that F contains the q0 th roots of unity when q is odd, and the 8th otherwise. Gu acts faithfully on a q0 -group Q such that Q/Z(Q) = U is the standard module and Z(Q) 1+2f m 1+2f m for odd q and Q ∼ is centralized by Gu , where Q ∼ = (q0 )+ = 20 otherwise. Let T be the standard holomorph to Q, and let ξ be a faithful irreducible character of T of degree q m . Then Q : Gu is a subgroup of T , and ResTGu (ξ) = ξu · µ where µ is the linear character of Gu of order gcd(q + 1, 2) (Appendix C3). Embed Z(Q) = Z(T ) into Z such that ξ and χ lie above the same linear character. (i) By assumption g = xz for some sth root of unity z ∈ Z and some x ∈ E, because Gu induces all diagonal automorphisms on E and |Gu /Su | = q + 1. Since Z(Su ) ⊆ Zu and s - |Zu | = q + 1, we may let hzi act trivially on T and regard ξ as a faithful character of T0 = T hzi with ξ(z)/ξ(1) = χ(z)/χ(1). (Either s = q0 and T = T0 or T0 = T × hzi.) Identify x with the (unique) element of Su of order s mapping onto x (so that g = xz gets an element of T0 ). Let ν be the linear character of hxi with ν(x) = χ(z) χ(1) . From Theorem 6.2c it follows that hξu , νihxi = hξ, 1ihgi ≤

1 m (q + (s − 1)q m−1 ) s

if s 6= q or s = 2 | q, and hξu , νihxi ≤ 2q m−1 otherwise. By hypothesis no eigenvalue 6= 1 of x on U has order dividing q + 1. Hence letting d0 be the dimension over Fq2 of CU (x), and letting dk = 0 otherwise, by Appendix (C3) ξj (x) =

¢ ½¡ q (−1)m X (−1)m (−q)d0 + q ¢ (−q)dk εkj = · ¡ (−q)d0 − 1 q+1 q+1 k=0

if j = 0 . if j > 0

Pq We just use that εj 6= 1 for j = 1, · · ·, q and that k=0 εjk = 0. We infer that the ξj (x) for j > 0 agree, and that ξ0 (x) − ξj (x) = (−1)m . There is a corresponding statement replacing x by any power xi 6= 1. The assertion Pq follows noting that ξu = j=0 ξj . (ii) Exclude (also) the case (m, q) = (6, 2) (where χ(g) = 0 or χ(g) = ±23 by [Atlas, p. 117]). Then some distinguished group Y appears in T , where Y = Gu .2 is a field extension group in the odd case and Y /Z(T ) ∼ = Gu .2 otherwise (Appendix C3) Since Out(L) is the semidirect product of


139

the cyclic groups of outer diagonal automorphisms (induced by Gu ) with that of field automorphisms, by assumption hL, Z(E)gi is isomorphic to a subgroup of Gu .2/Zu . So hE, gi is a section of Y . Observe that Gu /Su ∼ = Zu as modules for the group of field automorphisms. The outer field automorphism of order 2 inverts the elements of Zu = hzu i, because it sends zu to zuq = zu−1 . Consider first the case that q is even (hence q + 1 odd). By the construction of the irreducible Weil characters the outer field automorphism of Gu of order 2 leaves ξ0 invariant and fuses the other characters pairwise. Hence by assumption χ = ξ0 on E and E = L (as χ is faithful and Zu is the kernel of ξ0 ). Let g 0 ∈ Y map onto g. Then χ(g) = ξ0 (g 0 ), the character ξ0 suitably extended to Y . We have ξ(g 0 ) = ±ξ0 (g 0 ) as the ξj for j > 0 are fused pairwise by g 0 . Hence χ(g) = ±ξ(g 0 ). By Theorem 4.4 either ξ(g 0 ) = 0 or g 0 is good for U and ξ(g 0 )2 = |CU (g)|. Suppose χ(g) 6= 0. Then χ(g)2 = ξ(g 0 )2 = |CU (g 0 )| = |CU (g 0 z)| for all z ∈ Zu . Certainly CU (g 0 ) ∩ CU (g 0 zu ) = 0. Hence dim Fq CU (g 0 ) ≤ 1 1 a 2 dim Fq U = 2 2m = m and χ(g) = ±q for some integer a ≤ m/2. Let q be odd. Then the outer field automorphism of Gu of order 2 leaves ξ0 and ξ(q+1)/2 invariant and fuses the other irreducible Weil characters pairwise. Hence χ = ξ0 or χ = ξ(q+1)/2 on E. Let ξa , ξb denote the irreducible constituents of ResTY (ξ) having the kernels Zu = hzu i and hzu2 i, respectively. So ξa = ξ0 and ξb = ξ(q+1)/2 on Gu if m is even (Ker(µ) ⊇ Zu ), and vice versa otherwise. Also, E = L unless m is even, χ = ξb on E and q ≡ 3 (mod 4), in which case hE, gi is isoclinic to a subgroup of Y /hzu2 i. There is a 2-element g 0 of Y mapping onto Zu g (identified with Z(E)g). We have χ(g) = ±ξa (g 0 ) or ±ξb (g 0 ), possibly χ(g) = ±iξb (g 0 ) when E 6= L and the isoclinism is proper. By Theorem 4.4, g 0 is good for U (as q is odd). From Theorem 4.5a we infer that ξ(g 0 z) is rational for all z ∈ Zu , because g 0 z is a q00 -element. We have ξ(g 0 ) = ξ(g 0 z) = ξa (g 0 ) + ξb (g 0 ) when z is a square in Zu , and ξ(g z) = ξa (g 0 ) − ξb (g 0 ) otherwise. As before CU (g 0 ) ∩ CU (g 0 zu ) = 0. The map u 7→ u(1 + zu ) is an injection from CU (g 0 ) into CU (g 0 zu ), because ¡ ¢ u(1 + zu )g 0 = u(1 + zu−1 ) = u(1 + zu ) zu−1 0

for u ∈ CU (g 0 ). Similarly, the map v 7→ v(1 + zu−1 ) is an injection from CU (g 0 zu ) into CU (g 0 ). Hence dim Fq CU (g 0 ) = dim Fq CU (g 0 zu ) ≤ m. Moreover ξ(g 0 )2 = |CU (g 0 )| = |CU (g 0 zu )| = ξ(g 0 zu )2 and so ξ(g 0 ) = ±ξ(g 0 zu ) as

140


both are integers. It follows that ¡ ¢ ξa (g 0 ) + ξb (g 0 ) = ± ξa (g 0 ) − ξb (g 0 ) . Consequently ξa (g 0 ) = 0 or ξb (g 0 ) = 0. Thus if χ(g) 6= 0, then χ(g) = ±ξ(g 0 ) as both are integers, and χ(g)2 = |CU (g 0 )| = q 2a for some integer a ≤ m/2, as desired.

¤

Because of Proposition 7.8a we have to examine only some few cases. As before we proceed starting with the lowest rank of the groups involved. m = 3 : We have to consider the following. (i) L = U3 (3) and d ≤ 21: Suppose first χ = χ2 is the Weil character of (minimal) degree d = 6 [Atlas, p. 14]. This χ is rational-valued but extends to X = L.2 ∼ = G2 (2) requiring the 3rd roots of unity. If there is no regular orbit for X × Z, then 1 1 1 r6 ≤ |2A|(r4 + r2 ) + |3A|(2r3 ) + |3B|(3r2 ) + |7AB|(6r) + |2B|(2r3 ) 2 2 2 = 63(r4 + r2 ) + 56r3 + 1.008r2 + 2.592r + 504r2 . This implies that r < 11. Hence r = 5 (by coprimeness), and these case cannot happen for X × Z. So consider G = L × Z and r = 5. One checks that µ4χ,1 vanishes at r = 5, so there is no regular orbit for G. Also, the minimal point stabilizers in G are cyclic groups of order 4 which however get larger in G0 = NGL(V ) (L) = G.2. (Consider elements in the classes 4C and 8C of L.2). We show that there is v ∈ V such that CG0 (v) ∼ = S3 , as stated in Theorem 7.2a. Let x be an element of L in the class 3B. Then NL (hxi) ∼ = S3 has no fixed points on V ] , but x does. Let v be a nonzero vector fixed by x. Then CL (v) = hxi, and from Lemma 5.1b it follows that H = CG (v) ∼ = S3 . We assert that CG0 (v) = H. Otherwise H is of index 2 in this group, which forces that CG0 (v) ∼ = H × Z2 since Z(S3 ) = 1 and Aut(S3 ) = S3 . Here the generator of Z2 belongs to the conjugacy class 2B of L.2, whose elements do not centralize elements in the class 3A [Atlas]. Hence the assertion.


141

Let next χ = χ3 be the Weil character of degree 7 which extends (rationally) to X = L.2 = G2 (2). So G = G0 = X × Z. The counting argument yields regular vectors unless r < 8. For r = 5 we compute µ4χ,1 = (r − 5)(r5 + 6r4 + 31r3 − 159r2 − 1.760r + 2.457)/|X|, and we get some point stabilizer of order 2. For the algebraically conjugate (Weil) characters χ = χ4 , χ5 of L, which fuse in G2 (2), the counting argument again gives regular vectors unless r = 5. As before there is no regular orbit for G = G0 = L × Z4 when r = 5, and a point stabilizer of order 2. For all other irreducible characters of L = U3 (3) no prime order element of Aut(L) = G2 (2) has an eigenspace of dimension greater than d − 8, and from r8 ≤ 2|G2 (2)| it follows that r < 4. (ii) L = U3 (4) and χ is a Weil character: The character χ = χ2 of degree 12 extends to X = L.4, and for G = X × Z the counting method works. Similar statement for the other four Weil characters of L which fuse pairwise in L.2 and together in L.4. There are always regular orbits. (iii) L = U3 (5) and χ is a Weil character: In each case no noncentral element of Aut(L) = L.3 or of 3.L.3 of prime order has an eigenspace of dimension greater that d − 8. From r8 ≤ 2|3.L.3| it follows that r < 7. (iv) L = U3 (7) and χ is a Weil character: In the representations of degree at most 43 no prime element of Aut(L) = L.2 has an eigenspace of dimension greater than n − 18. From r18 ≤ 2|L.2| it follows that r < 3. m = 4 : By Proposition 7.8a we have to examine the following. (i) L = U4 (3) and d ≤ 45: Assume there is no regular orbit. For the 21-dimensional representations of Aut(L) = L.D8 (character χ = χ2 in the [Atlas, p. 54]) we then have r21 ≤ |2D|r15 + |2B|r14 + |2A|r13 + 2|Aut(L)| · r11 = 126r15 + 540r14 + 2.835r13 + 52.254.720r11 , which implies that r < 7, a contradiction. For the 35-dimensional representations of L.21 and L.22 no prime order element has an eigenspace of dimension greater than d − 10, and from r10 ≤ 2|Aut(L)| it follows that r < 6. The 20-dimensional representations of 2.L and 4.L are treated similarly. For the 15-dimensional representations of 31 .L.22 (characters χ = χ56 ) the counting method yields that r < 7, a contradiction. For the 36-dimensional

142


representations of 32 .L (or 32 .L.23 ) and 122 .L, and for the 45-dimensional representations of the former group(s), the same method applies. It remains to investigate the (faithful) 6-dimensional representations of X = 61 .L.22 (as a complex reflection group). The isoclinism type of X does not matter, and the character χ of X is an extension of one of the √ two algebraically conjugate characters χ72 (requiring −3) [Atlas]. The counting argument gives the existence of regular orbits for G = G0 = X ◦ Z when r ≥ 157. Now the table of marks for L = U4 (3) (and for L.21 ) is available in [GAP], and from that it is not difficult to compute the table of ¯ = L.22 (extension by a diagonal automorphism). One obtains marks for X that µ6χ,1 = (r − 13)(r − 19)(r − 25)(r − 31)(r − 37)/|L.22 |, and that µtχ,1 = µ6χ,1 for all multiples t of 6 dividing exp (X) = 23 · 32 · 5 · 7. So there are no regular orbits if and only if r ∈ {13, 19, 31, 37}. Similar computation leads to the minimal stabilizers as given in Theorem 7.2a. The two isoclinic variants of X = 61 .U4 (3).22 get isomorphic in G = X ◦ Z when r = 13 or r = 37, because then (r − 1)/6 is even. (ii) L = U4 (4) and χ = ξj is a Weil character: Here L = E = SU4 (4) is not an Atlas group, and Gu = GU4 (4) = L × Zu where Zu = hzu i is cyclic of order 5. Aut(L) = L.4 permutes cyclically the nontrivial elements of Zu = Z(Gu ), hence the Weil characters ξ1 , ξ2 , ξ3 , ξ4 (of degree 51). Let g ∈ G be a noncentral element of prime order s. If s 6= 5, by Lemma 7.8b (picking s = 2) dim F CV (g) ≤

1 4 (4 + (s − 1)43 ) ≤ 32. 5s

Let s = 5. Write g = xz where x ∈ L and z ∈ Z. Let ε = e2πi/5 , and let dk = dk (x) = dim F16 (zu−k x) for k = 0, · · ·, 4, where U is the standard P4 module for Gu . By Appendix (C3) ξj (x) = 51 k=0 (−4)dk εkj . Since x is faithful on U , dk ≤ 3 for each k. We conclude that the eigenspaces of x (and of g) on V are at most 13-dimensional when χ = ξ0 on L, and at most 12-dimensional otherwise. Let g ∈ G be an involution inducing an outer field automorphism on E = L. Then χ = ξ0 on L (of degree 52), and we consider the worst case G = G0 = L.4 × Z. By Lemma 7.8b either χ(g) = 0 or χ(g) = ±4a for 1 2 some integer a ≤ m 2 = 2. Hence dim F CV (g) ≤ 2 (52 + 4 ) = 34. From r18 ≤ 2 · |L.4| = 215 · 32 · 53 · 13 · 17


143

we obtain that r < 4. Hence there is a regular orbit, in all cases. m = 5 : We only have to examine the following. (i) L = U5 (2) and d ≤ 55: We have to deal with the three Weil characters ξ0 = χ2 (degree 10) and ξ1 = χ3 , ξ2 = χ4 [Atlas, p. 72]. For the other characters (χ5 of degree 44 and χ6 of degree 55), no prime order element of L.2 has an eigenspace of dimension larger than d − 16, and the result follows. Let us consider first χ = ξ0 . Suppose X = L.2 and G = G0 = X × Z. The usual counting argument yields that there are regular vectors unless r < 15. (The involutions in the class 2A have weak spectral pattern [8, 2], so one cannot improve the estimate without additional arguments. On the other hand, the counting argument yields the existence of strongly real vectors.) The only possibilities are r = 7 and r = 13 (by coprimeness). But √ the character ξ0 requires 2 −2 on L.2 outside L, and −2 is not a square mod 7 or 13. Hence we are left with the cases r = 7, 13, but we have to consider G = L×Z (and G0 = (L×Z).2). The table of marks for L = U5 (2) is in [GAP] (but not that for U5 (2).2). For U5 (2) we obtain that the only roots of the polynomials µ6χ,1 and µ12 χ,1 are 1, 5 and 7. So there is no regular orbit just when r = 7. In this case all polynomials µ6χ,H for subgroups H of order 2, 3 vanish at r = 7 but not for a Klein 4-group H ∼ = V4 . So there is v ∈ V such that CG (v) = H. We assert that CG0 (v) = H as well. From the [Atlas] we read off that two involutions in H must belong to the class 2B and one to 2A. If CG0 (v) 6= H, it is a dihedral group of order 8 containing an involution in the class 2C, which centralizes the involution in H in class 2A and interchanges the other ones. Notice that the elements in class 4D square to elements in class 2B and cannot exchange two involutions in different classes. But N (2A) is enlarged when passing from L to L.2 by replacing a split extension with 2.A4 on the top by a 2.S4 [Atlas], and for this one needs an element of order 8. Hence the assertion. For χ = ξj , j > 0, we have to consider G = L × Z. The counting argument yields regular vectors unless r = 7. Here the polynomial µ6χ,1 is irreducible over F7 . Hence there is a regular orbit. (ii) L = U5 (3) and χ is a Weil character: Here L is not an Atlas group. Let Gu = GU5 (3) = L × Zu where Zu = Z(Gu ) = hzu i has order 4. Recall that ξ0 (1) = 60 and ξj (1) = 61 for j > 0. Either G0 = L × Z and χ = ξ1 or ξ3 on L, or G0 = L.2 × Z and χ = ξ0 or ξ2 . Let G = G0 . Suppose g is

144


a noncentral element of prime order s in G. If s 6= 2, then by Lemma 7.8b (picking s = 3) dim F CV (g) ≤ b

1 5 1 (3 − (s − 1)34 )c ≤ b (35 + 2 · 34 )c = 33. 4s 12

d0 If g = xz is an involution, with x ∈ L and z ∈ Z, then ξj (x) = −1 4 ((−3) + (−3)d1 ij + (−3)d2 i2j + (−3)d3 i3j ) by the formula given in Appendix (C3). Here dk = dk (x) = dim F9 CU (zu−k x) (and i2 = −1). This yields that |χ(x)| ≤ 41 (34 + 32 + 2) = 23 and so dim F CV (g) ≤ 42.

Suppose that g is an involution inducing an outer field automorphism on L. Then by Lemma 7.8b either χ(g) = 0 or χ(g) = ±3a for some integer 5 a ≤ m 2 = 2 . So the eigenspaces of g on V are at most of dimension 30. 42 , and since r17 > 2|L.2| = Since for any g the fixed point ratio f (g, V ) ≤ 60 13 10 2 · 3 · 5 · 7 · 61 for r > 10, there exist regular orbits in each case. m = 6 : By the proposition we only have to examine L = U6 (2), with d ≤ 56. For the 56-dimensional representations of 2.L.2, the largest dimension of an eigenspace of a noncentral element of prime order is 40, and we obtain a regular orbit. So it remains to consider the Weil characters (ξ0 = χ2 of degree 22, the others corresponding to χ78 in the Atlas notation [Atlas, p. 116–121]. The counting method applies. m = 7 : By Proposition 7.8a we just have to examine the Weil characters of L = SU7 (2). Recall that Gu = GU7 (2) = L × Z3 and that ξ0 (1) = 42, ξj (1) = 43 for j > 0. Consider G = G0 , so either G = L × Z or G = L.2 × Z and χ = ξ0 on L. Let g ∈ G be a noncentral element of prime order s. If s ≥ 5, then for any possibility for χ = ξj by Lemma 7.8b dim F CV (g) ≤ b

1 7 (2 + (s − 1)26 )c ≤ 25. 3s

Let s = 3, and let g = xz for x ∈ L and z ∈ Z. Let ε = e2πi/3 and dk = dk (x) = dim F4 CU (zu−k x) for k = 0, 1, 2, ¡where U is the standard module ¢ d0 + (−2)d1 εj + (−2)d2 ε2j . for Gu . By Appendix (C3), χj (x) = −1 3 (−2) Using that ε + ε2 = −1 and that ξj (x) is an algebraic integer one obtains that ξ0 (x) ∈ {0, −3, 6, 9, 12, −21} and that ξj (x) ∈ {−2, 4, 10, 1 + 12ε, 1 + 12ε2 , −5 + 3ε, −5 + 3ε2 , 1 − 21ε, 1 − 21ε2 } for j = 1, 2. We deduce that no eigenspace of x (or g) on V has dimension greater than 21 when χ = ξ0 , and greater than 22 when χ = ξ1 or ξ2 .


145

Let s = 2. Consider first the case that g induces an outer field automorphism on L. Then χ = ξ0 on L, and by Lemma 7.8b either χ(g) = 0 or 7 χ(g) = ±2a for some integer a ≤ m 2 = 2 . It follows that dim F CV (g) ≤

1 (42 + 23 ) = 25 2

in this case. Let finally g be an involution inducing an inner or diagonal automorphism on L. We consider the worst case that χ = ξ0 on L, as before. From Lemma 7.8b it follows that dim F CV (g) ≤ 32. There is a unique conjugacy class 2A of unitary transvections t in Gu (which generates SUm (q); see [Aschbacher, 1986, (22.3) and (22.4)]). For this we have d0 (t) = dim Fq2 CU (t) = m − 1 = 6 and ξ0 (t) = −22 (C3). Hence if g = (−1)t, then indeed dim F CV (g) = 32. For the other noncentral involutions t1 , t2 in Gu , up to conjugacy, we have d0 (t1 ) = 5, χ(t1 ) = 10, and d0 (t2 ) = 4, χ(t2 ) = −6, so that the F -dimensions of their eigenspaces on V are at most 26. Hence, combining the previous estimates, there is a regular G-orbit on V provided r42 > 2|L.2|r26 + |2A|(r32 + r10 ) for r ≥ 13. This is true since 2|L.2| = 223 · 38 · 5 · 7 · 11 · 43 and |2A| = 2.709. Observing that the centre [U, t] of the transvection t consists of isotropic vectors, the size of the conjugacy class 2A is nothing but (|I(U )|−1)/(q 2 −1) where I(U ) is the set of isotropic vectors in U . As in Lemma 4.6b one computes |I(U )| = q 2m−1 + (−1)m (q m − q m−1 ). The remaining cases (Weil characters for SU8 (2) and SU9 (2)) are treated in the same manner. 7.9. Orthogonal Groups Let L be a simple orthogonal group with associated module of dimension m. Then the automorphism group and the Schur multiplier of L is known [Steinberg, 1967]. By [Guralnick–Saxl, 2003] the covering number c(L) ≤ m + 1, and c(L) ≤ m for m ≥ 5. Furthermore R0 (L) has been computed by [Tiep–Zalesskii, 1996]. On the basis of Lemma 7.3b this readily gives the following. Proposition 7.9. There exists a regular orbit except possibly when L ∼ = + ∼ Ω7 (3) and d ≤ 78, or L ∼ (2) and d ≤ 52, or L Ω (2) and d ≤ 84. = Ω− = 8 8

146


We have to examine these three cases and show that regular orbits exist except when L ∼ = Ω+ 8 (2), d = 8. Case 1: L = Ω7 (3) and d ≤ 78: We use [Atlas, pp. 106–108]. In the 78-dimensional representations of L.2, the largest eigenspace of an element of prime order is 56. Since r22 > 2|L.2| for r > 3 (even), we are done. For the 27-dimensional representations of 3.L the usual counting argument works as well. Case 2: L = Ω− 8 (2) and d ≤ 52: For the representations of L.2 of degree 34 and 51, the largest eigenspace of elements in the class 2D ([Atlas, p. 88]; |2D| = 136) has dimension at most d − 7, and no other element of prime order has an eigenspace of dimension greater than d − 12. From r12 ≤ 2|2D| + 2|L.2| it follows that r < 7. Hence the result. Case 3: L = Ω+ 8 (2) and d ≤ 84: For the characters of L [Atlas, p. 86] χ2 (degree d = 28, extendible to Aut(L) = L.S3 ), χ3 (d = 35, extendible to L.2), χ6 (d = 50, extendible to L.S3 ), and χ7 (d = 84, extendible to L.2) observe that the elements in the class 2F have no eigenspaces of dimension greater than d − 7. All other elements of prime order have no eigenspaces of dimension greater than d − 12, which holds also for the characters of the above degrees which do not split in a proper extension of L. Since r28 > |2F |(r21 + r7 ) + 2|L.S3 |r16 for r > 7, there is a regular orbit. Similarly, for the 56-dimensional representations of 2.L.2 none of the noncentral elements of prime order has an eigenspace of dimension greater than 35, and the result follows. It remains to consider some faithful character χ = χ54 of degree 8 of E = 2.L, which is extendible to any X = 2.L.2. The multiplier M(L) is elementary of order 4 but Aut(L) permutes its involutions. The isoclinic type being irrelevant it suffices to consider X = W (E8 ), the Weyl group of the root lattice E8 . The spectral pattern of an involution g in the class 2F0 is gV = [1(7) , −1(1) ], and |2F | = 120. So it is clear that the counting method cannot work. (We only get in this manner regular vectors when ¯ = X/Z(X) is available in [GAP]. One r > 275.) The table of marks for X gets: µ2χ,1 = (r − 7)(r − 11)(r − 13)(r − 17)(r − 19)(r − 23)(r + 91)/|L|, µ4χ,1 = µ2χ,1 − (r − 13)(r − 17)(r − 29)/11.520, µ6χ,1 = µ2χ,1 − (r − 7)(r − 13)(r − 19)/38.880,


147

2 µ10 χ,1 = µχ,1 − (r − 11)/150, 6 µ12 χ,1 = µχ,1 − (r − 9)(r − 13)(r − 37)/11.520, 4 µ20 χ,1 = µχ,1 − (r − 41)/150, 6 µ30 χ,1 = µχ,1 − (r − 31)/150, and t µ7t χ,1 = µχ,1 −

1 7

for all positive even integers t.

One concludes that G = G0 = X ◦ Z has no regular orbit on V if and only if r ∈ {11, 13, 17, 19, 23}. Further inspection gives the minimal stabilizers listed in Theorem 7.2a. The existence of a (strongly) real vector can be seen also as follows. The representation of X of degree 8 comes from the action of X as automorphism group on the 8-dimensional root lattice E8 . X is transitive on the 240 minimal vectors, and a point stabilizer is isomorphic to the real group S6 (2) × Z2 . 7.10. Exceptional Groups Let L be a simple exceptional group of Lie type (including the twisted Suzuki and Ree groups). Then Aut(L) and M(L) are known. By [Guralnick and Saxl, 2003] the covering number c(L) ≤ ` + 3, where ` is the untwisted rank of L, except possibly when L = F4 (q) with c(L) ≤ 8. From [Seitz– Zalesskii, 1993] one has lower bounds for R0 (L). On the basis of Lemma 7.3b this yields the following. Proposition 7.10. There are regular orbits except possibly when L = G2 (3) and d ≤ 27, or L = G2 (4) and d = 12, or L = Sz(8) and d ≤ 56, or L =3 D4 (2) and d ≤ 52, or L =2 F4 (2)0 and d ≤ 52, or L = F4 (2) and d = 52. The remaining groups to be examined are all Atlas groups, and the counting method applies in each case. This completes the proof of Theorem 7.2a. At this stage, the k(GV ) problem is settled in all characteristics p different from 3, 5, 7, 11, 13, 19 and 31. This is a consequence of the classification of the nonreal reduced pairs in Theorems 6.1, 7.1, in terms of the Robinson–Thompson criterion (Theorem 5.2b) and Clifford reduction (Theorem 5.4). The challenge now is to describe effectively the pairs (G, V ) admitting no real vectors, which must “involve” nonreal reduced pairs in some Clifford-theoretic sense.

Chapter 8

Modules without Real Vectors In solving the k(GV ) problem we may assume that G is irreducible on V (Proposition 3.1a) and that no vector in V is real for G (Theorem 5.2b). We show that then (G, V ) must be a “nonreal induced” pair, that is, obtained by module induction from a nonreal reduced pair. This is an important step towards the solution of the problem. 8.1. Some Fixed Point Ratios As usual F = Fr is a finite field of characteristic p not dividing the order of the finite group G, and V is a F G-module. Let Z ∼ = F ? be the group of scalar multiplications on V . Lemma 8.1a. Suppose V = W1 ⊗F · · · ⊗F Wn is a tensor decomposition into F G-modules. Let gi ∈ GL(Wi ) and g = g1 ⊗ · · · ⊗ gn in GL(V ). (i) If f (zgi , Wi ) ≤ m for all z ∈ F ? and all i, then f (g, V ) ≤ m. (ii) Let v = w1 ⊗ · · · ⊗ wn for nonzero vectors wi ∈ Wi . If G induces all scalar transformations on the Wi , then CG (v)/CG (V ) is isomorphic to a subgroup of CG (w1 )/CG (W ) × · · · × CG (wn )/CG (W ). Proof. (i) If some gi is a scalar transformation on Wi , say with zi , then f (zi−1 gi , Wi ) = 1 and so the result is obvious (m ≥ 1). So exclude this. We argue by induction on n, the statement being obvious for n = 1. So let n > 1, and let the εj ∈ Z be the distinct eigenvalues of g1 on W1 . Then f (g, V ) =

X

f (ε−1 j g1 , W1 ) · f (εj g2 ⊗ g3 ⊗ · · · ⊗ gn , W2 ⊗ · · · ⊗ Wn ) ≤ m

j

by induction since

P j

f (ε−1 j g1 , W1 ) ≤ 1.

(ii) Let g ∈ CG (v), and let gi ∈ GL(Wi ) be induced by g. Then wi g = wi gi = zi wi for some zi ∈ Z, and z1 · · · zn = 1. By hypothesis zi−1 gi is induced on Wi by some element in G (centralizing wi ). Hence the assignment g 7→ (z1−1 g1 , · · ·, zn−1 gn ) is a homomorphism from CG (v) into 148

Modules without Real Vectors

149

CG (w1 )/CG (W1 ) × · · · × CG (wn )/CG (Wn ). Its kernel is CG (V ), because if gi = zi on Wi for each i, then g is the identity on V . ¤ Lemma 8.1b. Let X = Hwr Sn for some finite group H and some integer n ≥ 2. Suppose V = W ⊗n for some faithful F H-module W of dimension d = dim F W ≥ 2, with X acting on V as usual (1.2). Let N = H (n) be the base group of the wreath product. Then f (x, V ) ≤ 1− d1 for x ∈ N rCX (V ), 1 for x ∈ X rN . Moreover, if x ∈ X rN is a p0 -element, and f (x, V ) ≤ 1− 2d d+1 then f (x, V ) ≤ 2d . Proof. Since f (h, W ) ≤ (d − 1)/d = 1 − d1 for each h ∈ H ] , the first statement follows from Lemma 8.1a. Let x ∈ X r N , and observe that CX (V ) ⊆ N . There exists y ∈ N such that [x, y] 6∈ CX (V ). We have dim F CV (x) = dim F CV (x−1 ) = dim F CV (y−1 xy) and CV (x−1 )∩CV (y −1 xy) ⊆ CV ([x, y]). Hence 2 dim F CV (x) ≤ dim F V + dim F CV ([x, y]) ≤ dim F V(1 + (1 − as [x, y] ∈ N r CX (V ). Hence f (x, V ) ≤ 1 −

1 )) d

1 2d . 0

Suppose finally that x ∈ X r N is a p -element and N x ∈ Sn is a product of r disjoint cycles. Let χ be the Brauer character of X afforded by V , and let θ be that of H afforded by W . Then χ = TenX Y (θ) where Y is the stabilizer in X of some tensor factor W of V (Y /N ∼ = Sn−1 ). By the character formula (1.2e), χ(x) is the product of r ≤ n−1 values of θ. Hence P |χ(x)| ≤ θ(1)r ≤ χ(1)/d. Now dim F CV (x) ≤ |χ(x)| + i dim F CV (z−1 i x), the sum taken over all eigenvalues zi 6= 1 of x on V , and dim F CV (x) + P −1 i dim F CV (zi x) ≤ dim F V = χ(1). This gives the asserted estimate 1 ¤ 2f (x, V ) ≤ d + 1. 8.2. Tensor Induction of Reduced Pairs Recall the Frobenius embedding of a group into a wreath product (Theorem 1.2a). We need a slight improvement. Lemma 8.2a. Let Y be a subgroup of the group X with index n. Assume that there is a normal subgroup C of Y such that CoreX (C) = 1. Then there is an embedding of X into Y¯ wr Sn where Y¯ = Y /C. In the discussion of Theorem 1.2a replace the elements xi ∈ Y by their cosets Cxi .

150


Proposition 8.2b. Suppose V = TenG H (W ) where (H/CH (W ), W ) is a nonreal reduced pair of quasisimple type. If H is a proper subgroup of G, then there is a strongly real vector in V for G. Proof. As usual V is a coprime F G-module. We know from Theorems 7.1 and 7.2 that F = Fp with p ∈ {7, 11, 13, 19, 31} and that d = dim F W = 2 or 4. We may assume that H/CH (W ) = E ◦ Z is large, that is, Z ∼ = F ?. ± Here E is one of 2.A5 , 2 S6 or Sp4 (3). There exist pairs which are not large only when E = Sp4 (3) and p = 13 or 19. But we may certainly assume that G induces all scalar multiplications on V = W ⊗n (see the comment after 5.3b). If g ∈ G induces the scalar c on V , it preserves each tensor factor and induces certain scalars ci , with c = c1 · · · cn . Using that F ? is cyclic we see that H induces all scalar transfomations on W . Let n = |G : H|. By hypothesis n ≥ 2. By Lemma 8.2a there is a (Frobenius) embedding of G/CG (V ) into T = (EZ))wr Sn . We regard G/CG (V ) as a subgroup of G0 = T /CT (V ) (the wreath product acting in the obvious way). Then G0 = X ◦ Z where X is the image in G0 of Ewr Sn . Let N and N0 be the images in G0 of the base groups of Ewr Sn and T , respectively. Then N ⊆ X and Z(N ) = Z(X) ∼ = Z(E) has order 2, and N0 = N ◦ Z. Of course ∼ G0 /N0 ∼ X/N S = = n . The elements of N will be written as n-tuples (h1 , · · ·, hn ) with hi ∈ E (and the obvious central amalgamation). Let Y be the stabilizer in X of some tensor factor W of V , so that Y /N ∼ = Sn−1 . Let χ be the Brauer character of X afforded by V , and let θ be that of E afforded by W . Then χ = TenX Y (θ) as in (1.2e). We sometimes view θ as a (faithful) character of N0 = N ◦ Z and χ as a character of G0 = X ◦ Z. n

We have |W | = pd and |V | = pd . Let f be the maximum of the f (x) = f (x, V ) taken over all noncentral x ∈ X. Since |X/N | = |Sn | = n! and |N/Z(N )| = |E/Z(E)|n , by (5.6d) there is a regular G0 -orbit on V , hence a regular G-orbit, provided n!|E/Z(E)|n (pbf d

n

c

+ pd

Now we examine each possible case.

n

−bf dn c

n

) < pd = |V |.


151

Type (2.A5 ) : Here E ∼ = 2.A5 ∼ = Sp2 (5), θ is a (faithful) Weil character of E of degree d = 2, and p = 11, 19 or 31 (Sec. 7.1). Also |E/Z(E)| = 60, and f ≤ 43 by Lemma 8.1b. There is a regular G-orbit on V if n!60n < n−2

p2 1+1/p2n−1

. This holds true for n ≥ 6, and all p. It remains to investigate

the cases n ≤ 5. Then G0 is a p0 -group, and we may take G = G0 . We use that E, Ewr Sn and, therefore, X are real groups. Let p = 31. Then G = X × Z15 . The above estimate also holds for n = 5. By Lemma 8.1b we have f (g) ≤ 21 if g ∈ N is noncentral. Let n = 4. Then all 5-elements of G are in N0 , and if x ∈ X r N has order 3, then N x ∈ S4 has two cycles (orbits) and so |χ(x)| ≤ θ(1)2 = d2 = 4. It follows that x has the eigenvalue 1 on V with multiplicity at most 8, because the primitive 3rd roots of unity have real part − 21 (and 9 − 7/2 > 4). Thus 1 f (x) ≤ 21 , as before. Now |G| = 24 · 604 · 30 < 318 = |V | 2 , whence 1 |G| · |V | 2 < |V |. We infer that there is v ∈ V such that CG (v) contains no elements of order 3 or 5, that is, CG (v) is a 2-group. Thus CG (v) ⊆ X and v is strongly real for G. Let n = 3 (and p = 31). For x ∈ X r N of order 3 we have |χ(x)| = θ(1) = 2 (as N x is a 3-cycle). Hence xV = [4, 2, 2]. Fix an element h ∈ E of order 5, and let g = (h, h, h) in N . Then CX (g) has order 2 · 53 · 6, and CX (g) contains x. We infer that there are at most 2 · 53 = 250 subgroups in X of order 3 outside N . There are 203 + 3 · 202 + 3 · 20 = 2 · 4.630 elements y ∈ N of order 3, all satisfying χ(y) = ±1 (as θ(3A0 ) = −1). We get yV = [3, 3, 2]. There are 243 + 3 · 242 + 3 · 24 = 4 · 3.906 elements of order 5 in N , all having eigenvalues on V with multiplicity at most 4 by Lemma 8.1b. Since 250(p4 + 2p2 ) + 4.630(2p3 + p2 ) + 3.906(2p4 ) < p8 = |V |, there exists v ∈ V such that CG (v) is a 2-group. We are done as before. The n = 2 case is treated similarly. Let p = 19. Then every 30 -subgroup of G is a real group. Let g ∈ G be of order 3. If g ∈ N0 , then f (g) ≤ 21 by Lemma 8.1b. If g 6∈ N0 then its image in Sn ∼ = G/N0 is a product of at most n − 2 disjoint cycles. Consequently the real part Re(χ(g)) ≤ |χ(g)| ≤ θ(1)n−2 = 22 (b)

(c)

n−2

.

Let gV = [1(a) , z1 , z2 ] where z1 6= z2 are the primitive 3rd roots of unity in F . Then f (g) = a/2n and Re(zi ) = − 21 . For n = 5 we have Re(χ(g)) ≤ 8,

152


which implies that a ≤ 16 and f (g) ≤ 21 . Similarly a ≤ 8 for n = 4, hence n 1 f (g) ≤ 21 as before. For n = 4, 5 we have |G| = n!60n · 18 < 192 /2 = |V | 2 1 and so |G| · |V | 2 < |V |. We deduce that there is v ∈ V such that CG (v) is a 30 -group, as desired. The case p = 19, n = 3 is handled by showing that NX (F v) is a 30 group for some v ∈ V . By Lemma 5.1b then v is strongly real for G. There are 2 · 103 elements of order 3 in N . Since θ(h) = −1 for each h ∈ E of order 3 [Atlas], for any y ∈ N of order 3 we have χ(y) = ±1 and necessarily yV = [3, 3, 2] (so χ(y) = −1). As before we see that there are at most 4 · 303 subgroups of order 3 in X outside N (where we know the spectral pattern already). Since 250(p4 + 2p2 ) + 4.630(2p3 + p2 ) < p8 = |V |, there is v as required. In the n = 2 case we pick an element x = (h, h) in X of order 5 (h ∈ E). If T is a subgroup of X containing x, then NX (T )/T is a 30 -group. The character table of E ∼ = 2.A5 gives that dim F CV (x) = 2. Now let v be any nonzero vector in CV (x), and let T = CX (v). By Lemma 5.1b this v is strongly real for G0 . Let p = 11. Then G = X × Z5 . Each 50 -subgroup of G is real. For n = 5, 4 we claim that there is v ∈ V such that CG (v) is a 50 -group. Let g ∈ G be an element of order 5. If g ∈ N0 then f (g) ≤ 21 by Lemma 8.1b. Otherwise n = 5 and N0 g is a 5-cycle in S5 . Then |χ(g)| = θ(1) = 2 and dim F CV (g) = 8 as each primitive 5th root of unity in F appears as eigenvalue of g on V with multiplicity 6. Hence f (g) = 41 in this case. 1 Now use that |G| · |V | 2 < |V |, which gives the claim. Let n = 3, 2, and let y = (h, h, h) resp. y = (h, h) for an element h ∈ Y of order 3. Then χ(y) = θ(h)n = (−1)n , and we deduce that dim F CV (y) = 2 in both cases. Let v be a nonzero vector in CV (y), and let C = CX (v). Then NX (C)/C is a 50 -group. Now use that χ takes only real values on X, and use Lemma 5.1b. This shows that v is a strongly real vector for G. Type (2.A6 ) : Here E ∼ = 2± S6 , d = 4 and p = 7. By Lemma 8.1b, 7 f ≤ 8 . We obtain that there is a regular G0 -orbit on V provided n ≥ 6. So let n ≤ 5. Then G0 is a p0 -group, and we take G = G0 . By Lemma 8.1b now f ≤ 43 . This yields the existence of a regular G-orbit on V for n ≥ 3. So let n = 2. Inspection of the character table [Atlas, p. 5], noting that θ = χ9 for E = 2+ S6 and θ = χ8 for E = 2− S6 , yields that f (g) ≤ 21 for each noncentral element g ∈ G of order 3. (If g ∈ N either


153

χ(g) = 1 and gV = [6, 5, 5], or χ(g) = −2 and gV = [4, 6, 6], or χ(g) = 4 and gV = [8, 4, 4].) The number of noncentral elements of order 3 in G is 1 at most 3 · (802 + 2 · 80) < 78 = |V | 2 . It follows that there is v ∈ V such that CG (v) is a 30 -group (contained in N ). The restriction to CG (v) of χ is rational-valued [Atlas]. ∼ Sp (3), d = 4 and p ∈ {7, 13, 19} (Sec. 7.1). Type (Sp4 (3)) : Here E = 4 Recall also that θ = χ21 or χ22 in the notation of the Atlas [Atlas, p. 27]. From the general estimate we get the existence of a regular G-orbit on V when p = 13 or 19 and n ≥ 3, and when p = 7 and n ≥ 4. Let p = 19 and n = 2. Then G0 = X × Z9 is a p0 -group. Each element of order 3 in G0 lies in N0 , and the number of such elements is at most 1 |G0 |/2 < 198 = |V | 2 . Hence there is v ∈ V such that CG0 (v) is a 30 -group. Then the restriction to CG0 (v) of χ is rational-valued [Atlas]. Let p = 13 and n = 2. Then G0 = X × Z6 is a p0 -group. Let G = G0 . Each element in X of order 3 lies in N . If x = (h, 1) or (1, h) with h ∈ E belonging to one of the conjugacy classes 3A0 B0 , then xV = [12, 4] . There are 2·80 such elements. For the remaining 802 +2402 +2·240+4802 +2·480 = P 2 · 147.920 elements y of order 3 in X we have z∈Z |CV (zy)| ≤ p9 + p6 + p. It follows that X

|CV (γ)| ≤ 80(p12 + p4 ) + 147.920(p9 + p6 + p).

γ∈β3 (G)

Let g ∈ N be a noncentral involution. Then g = (h1 , h2 ) where either both hi ∈ E belong to the class 2A0 or one hi lies in 2A0 and the other is 1. Then χ(g) = 0 [Atlas] and so gV = [8, 8]. There are 452 + 2 · 45 = 2.115 P such involutions in N . It follows that γ∈β2 (G) |CV (γ)| ≤ 2.115(2p8 ). We P have γ∈β2,3 (G) |CV (γ)| < |V | = 1316 . We conclude that there is v ∈ V such that CG (v) = CX (v) is a {2, 3}0 -group. The restriction to CX (v) of χ is rational-valued [Atlas]. P Let p = 7 and n = 2, 3. As before we prove that γ∈β3 (G0 ) |CV (γ)| < n |V | = 72 . There is v ∈ V such that CG0 (v) = CX (v) is a 30 -group, and the restriction to CX (v) of χ is rational-valued [Atlas]. We are done. ¤ Proposition 8.2c. Suppose V = TenG H (W ) is a coprime F G-module where (H/CH (W ), W ) is a reduced pair of quasisimple type. There is v ∈ V such that CG (v)/CG (V ) has a regular orbit on V .

154


Proof. We may and do assume that H induces all scalar transformations on W . Let d = dim F W and let n = |G : H|. In view of Corollary 7.2c we may assume that n ≥ 2. Pick w ∈ W such that C = CH (w)/CH (W ) is a point stabilizer of minimal order. That is, either C = 1 or C is as listed in Theorem 7.2a, or (H/CH (W ), W ) is a permutation pair (Example 5.1a). In this latter case either C = 1 or r = p = d + 2 or d + 3, in which case we may choose w such that C is cyclic of order dividing p − 1. In each case let v = w ⊗ · · · ⊗ w ∈ W ⊗n = V. Let N = CoreG (H). So S = G/N is a transitive permutation group of degree n, which is a p0 -group (like G). By Lemma 8.1a, CN (v)/CN (V ) is isomorphic to a subgroup of C (n) . It follows that CG (v)/CG (V ) is isomorphic to a subgroup of Cwr S. In particular, |CG (v)/CG (V )| is a divisor of |C|n · n!. Let F = Fr , so that n |W | = rd and |V | = rd . Suppose first that d ≥ 3. Then f (g) = f (g, V ) ≤ 1 − d1 for each g ∈ GrCG (V ) by Lemma 8.1b. Hence the number of vectors in V centralized by 1 some nontrivial element in CG (v)/CG (V ) is at most equal to n!|C|n |V |1− d . There is a regular vector in V for CG (v) provided this number is less than |V |, that is, if n−1 n!|C|n < rd . This is fulfilled for all n ≥ 2 when C = 1 or when |C| ≤ p − 1 (handling the cases where (H/CH (W ), W ) is a permutation pair). Otherwise r = p and |C| ≤ 72 by Theorem 7.2a. Then the above inequality is fulfilled for all n if d ≥ 5. For d = 3 we even have p ≥ 11 and |C| ≤ 5, which gives the result. Let d = 4. Then the above inequality holds for all n ≥ 3. So let n = 2. If |C| ≤ 18, the result holds. Otherwise by Theorem 7.2a, H/CH (W ) ∼ = Sp4 (3)×Z3 , d = 4, r = 7, and by Corollary 7.2c there exists a regular vector w e1 ∈ W ] for C ∼ e1 i ∩ w e1 C = {w e1 }. = SL2 (3) × Z3 such that hw Let w e2 ∈ W be any vector linearly independent from w e1 . Then w e1 ⊗ w e2 is a regular vector for Cwr S2 and hence for CG (v)/CG (V ). Let d = 2. Then f (g, V ) ≤ 43 for each nontrivial element in G/CG (V ) by Lemma 8.1b. There is a regular CG (v)/CG (V )-orbit on V provided n!|C|n < r2

n−2

.

Since r ≥ 3 is odd, this inequality holds when C = 1. So let C 6= 1, in which case H/CH (W ) ∼ = 2.A5 ◦Z by Theorem 7.2a. If r 6= 11, then |C| ≤ 3,


155

and the inequality holds for all n ≥ 2. So let F = F11 and |C| = 5. Then the estimate holds for n ≥ 5. Let n = 4. We have |P1 (W )| = 12, and C has on P1 (W ) two fixed points F w = F w1 and F w2 , say, and two orbits of length 5 represented by F w3 and F w4 (say). The vector w1 ⊗ w2 ⊗ w3 ⊗ w4 is regular for Cwr S4 and hence for CG (v)/CG (V ). The cases n = 3, 2 are treated similarly. ¤ 8.3. Tensor Products of Reduced Pairs Proposition 8.3a. Suppose V = U ⊗F W where both (G/CG (U ), U ) and (G/CG (W ), W ) are reduced pairs. Then there is a real vector in V for G. Proof. Without loss we may assume that G induces on U and on W all scalar transformations. By Proposition 5.3b we may also assume that one of the reduced pairs is nonreal, say the first one. Then d = dim F U = 2, 3 or 4 and r = p ∈ {3, 5, 7, 11, 13, 19, 31} by the classification of the nonreal reduced pairs (F = Fr ). By Theorems 6.1 and 7.1, and by Corollary 7.2c, there is w ∈ W such that CG (w)/CG (W ) has a regular orbit on W . Hence the assertion follows from part (iii) of Proposition 5.3b provided dim F W ≥ d. But when dim F W < d, we may exchange the roles of U and W . ¤ f f f Proposition 8.3b. Let W = TenG H (W ) where (H/CH (W ), W ) is a reduced pair of quasisimple type. Suppose V = U ⊗F W where (G/CG (U ), U ) is a reduced pair. Then there is a real vector in V for G. Proof. Without loss of generality we may assume that G induces all scalar transformations on U and on W . If H = G the result follows from Proposition 8.3a. So let n = |G : H| ≥ 2. Combining Propositions 8.2b and 5.3c we see that there is a real vector in W for G, at any rate. If, in addition, there is a real vector in U for G, the result follows from part (i) of Proposition 5.3b. Hence we may assume that (G/CG (U ), U ) is a nonreal reduced pair. By the results in Chapters 6, 7 then dim F U = 2, 3 or 4. On the other f ≥ 2 and so dim F W = ( dim F W f )n ≥ 4. By Proposition hand, dim F W 8.2c there is w ∈ W such that CG (w)/CG (W ) has a regular orbit on W . Therefore part (iii) of Proposition 5.3b applies again. ¤ Proposition 8.3c. Let V = U1 ⊗F U2 ⊗F U3 where the (G/CG (Ui ), Ui ) are nonreal reduced pairs. Then there exists a real vector in V for G.

156


Proof. We may assume that G induces all scalar transformations on the Ui . Suppose first that there are abelian vectors ui ∈ Ui for G at least two times, say for the indices i = 2, 3. Let W = U2 ⊗F U3 and let w = u2 ⊗ u3 . By part (ii) of Lemma 8.1a, w then is an abelian vector for G/CG (W ). Since dim F W ≥ 4 ≥ dim F U1 by Theorems 6.1, 7.1, the assertion follows from part (iii) of Proposition 5.3b. It remains to examine the situations where there are no such abelian vectors. By Theorems 6.1, 7.1 then at least two of the pairs are of type (Sp4 (3)), say the latter ones, and p = 7 or p = 13. Then U = U2 = U3 and G acts on W = U2 ⊗F U3 = U ⊗2 like the base group of (G/CG (U ))wr S2 . Hence the result follows from Proposition 8.3b. ¤ 8.4. The Riese–Schmid Theorem We call a pair (G, V ) nonreal induced provided V is a faithful coprime F Gmodule which admits no real vector for G and which is induced from some nonreal reduced pair (H, W ), that is, V = IndG G0 (W ) for some subgroup ∼ G0 of G satisfying H = G0 /CG0 (W ). From the classification of the nonreal reduced pairs it follows that then V is an absolutely irreducible F G-module and F = Fp for one of the primes p = 3, 5, 7, 11, 13, 19 or 31. Of course this also forces that |G : G0 | is not divisible by p. Theorem 8.4 (Riese–Schmid). Let V be a faithful, irreducible, coprime F G-module admitting no real vector for G. Then (G, V ) is nonreal induced. Proof. Assume V is a counterexample with d = dim F V minimal. So there is no real vector in V for G, hence G is nonabelian and d ≥ 2. Furthermore, whenever V0 is an irreducible coprime F0 G0 -module for which char(F0 ) = p = char(F ) and dim F0 V0 < d, then there is a real vector in V0 for G0 , that is, for G0 /CG0 (V0 ), or (G0 , V0 ) is nonreal induced. We argue, at first, like for Theorem 5.4. (1) V is an absolutely irreducible F G-module: For otherwise embed F (properly) into the field F0 = EndF G (V ). Then F0 ⊗F V is the direct sum of the distinct Galois conjugates over F of some absolutely irreducible F0 G-module V0 . This V0 is a faithful module and dim F0 V0 < d. Thus either V0 contains a real vector v0 for G, or V0 = IndG H (W ) where (H/CH (W ), W ) is a nonreal reduced pair (over F0 ). In the former case the sum of the distinct Galois conjugates over F of v0 is


157

a real vector in V for G. The second case cannot occur by the classification of the nonreal reduced pairs, because F0 is not a prime field. (2) V is a primitive F G-module: Otherwise V = IndG H (U ) for some proper subgroup H of G and some F Hmodule U . Then dim F U < d. Hence either there is a real vector in U for H, which implies that there is a real vector in V for G by part (i) of Proposition 5.3a, or U = IndH Y (W ) where (Y /CY (W ), W ) is a nonreal reduced pair. By transitivity of module induction V = IndG Y (W ) in the latter case. (3) The irreducible constituents of ResG N (V ) are absolutely irreducible for all normal subgroups N of G: Otherwise choose N maximal such that (3) is false. Then N 6= G by (1). As in the proof for Theorem 5.4 one gets that ResG N (V ) = W is an irreducible F N -module with F0 = EndF N (W ) being a proper extension field of F . Moreover G/N ∼ = Gal(F0 |F ) and F0 ⊗F V ∼ = IndG N (U ), where U is an (absolutely) irreducible constituent of F0 ⊗F W . If there is a real vector in U for N , then there is a real vector in F0 ⊗F V for G by Proposition 5.3a, and we get a real vector in V . Hence by the choice of V we have U = IndN Y (U0 ) for some subgroup Y of N and some F0 Y -module U0 , with (Y /CY (U0 ), U0 ) being a nonreal reduced pair. But such a pair does not exist. Hence the assertion. From (1), (2), (3) it follows that every abelian normal subgroup of G is cyclic and central in G (acting by scalar multiplications). The generalized Fitting subgroup of G is nonabelian for otherwise G were cyclic and so had a regular orbit on V . (4) There is a nonabelian normal subgroup N of G such that ResG N (V ) is not (absolutely) irreducible: Assume the contrary. Let E be a minimal nonabelian normal subgroup of G. By assumption and (2), (3), ResG E (V ) = W is absolutely irreducible. Hence CG (E) = Z = Z(G) acts via scalar multiplications on V and EZ is the generalized Fitting subgroup of G. Suppose first that E is solvable. Then it is a q-group of “symplectic type” for some prime q 6= p. Either q is odd and E = Ω1 (EZ) is of exponent q, or E is an extraspecial 2-group

158


or the central product of such one with a cyclic group of order 4 (when F contains the 4th roots of unity). Hence (G, V ) is a nonreal reduced pair with core E, a contradiction. Suppose next that E is not solvable. If E is quasisimple, we get a contradiction as before. Hence E is the central product of n ≥ 2 distinct G-conjugates of some quasisimple group E0 . Let G0 = NG (E0 ), so that n = |G : G0 |. Then W is the tensor product of n distinct G-conjugates of the unique absolutely irreducible constituent W0 of ResE E0 (W ). Let θ and θ0 be the Brauer characters of E, E0 afforded by W , W0 , respectively. By c0 extending W0 (in the usual Theorem 1.9c there¡ is an F G0 (θ0 )-module W ¢ c0 ), W c0 is reduced with quasisimple sense). The pair G0 (θ0 )/CG0 (θ0 ) (W b of core isomorphic to E0 . By Theorem 1.9d there is a finite extension G 0 b G = G(θ) by an abelian p -group, containing a subgroup G0 mapping onto G0 (θ0 ), such that b (W c V = TenG b0 0 ). G c0 for G b 0 , there is a real vector for G in V by If there is a real vector in W c0 for G b 0 , by Proposition Proposition 5.3c. If there is no real vector in W 8.2b there is a real vector in V for G as well. This proves (4). e of G we have V = U ⊗F W where (5) Over some central extension G e the pair (G/CG e(U ), U ) is nonreal reduced and where W is an absolutely e irreducible and primitive F G-module with dim F W ≥ 2: By (4), (3) there is a nonabelian normal subgroup N of G such that the restriction of V to N is a proper multiple eU of some absolutely irreducible F N -module U (e ≥ 2). This U is faithful. Now argue as in the proof for Theorem 5.4, on the basis of stable Clifford theory. This gives the tensor decomposition V = U ⊗F W as asserted where dim F U ≥ 2 and dim F W ≥ 2. By (1), (2) both U and W are absolutely irreducible and primitive. For ¢ e (W e ¡ResG e (U f ) were imprimitive, say, then V = IndG e f if W = IndG e e e ) ⊗F W H H H and (2) applies. Since both dim F U and dim F W are less than d, the theorem holds for U and for W . At least one of these modules does not contain a real vector by Proposition 5.3b, hence defines a nonreal reduced pair by the choice of V . The situation being symmetric we assume that e e (G/C e(U ), U ) is a nonreal reduced pair. G (6) We have F = Fp for a certain odd prime p ≤ 31. There is a real e and (G/C e vector in W for G, e(W ), W ) is neither reduced nor is obtained G by (proper) tensor induction from a reduced pair with quasisimple core:


159

The first statement follows from (5) by the classification of the nonreal reduced pairs. Apply furthermore Propositions 8.3a and 8.3b. Now we proceed by showing that statements (3), (4) hold for W in place of V . e (W ) are absolutely irreducible (7) The irreducible constituents of ResG e N e of G: e for all normal subgroups N e of G e which is maxiOtherwise there exists a proper normal subgroup N e (W ) has an irreducible sumf = ResG mal subject to the property that W e N f is irreducible, F0 = mand which is not absolutely irreducible. Then W f e e ∼ EndF Ne (W ) is a proper extension field of F and G/N = Gal(F0 |F ) = Γ f (as before). Furthermore, denoting by W0 an (absolutely) irreducible cone (W f , we have F0 ⊗F W ∼ f e stituent of F0 ⊗F W = IndG e(V ). Then e 0 ). Let Z = CG N ? e e e G = G/Z, as G is faithful on V , and Z induces scalars from F on both U e therefore Ze ⊆ N e . Let N = N e /Z. e and W . By the (maximal) choice of N ∼ e e e Since G/N = G/N is cyclic, G/CG e(U ) contains a normal subgroup mape G e e ping onto the core of G/C e(U ). It follows that U = ResNe (U ) is absolutely G e0 = F0 ⊗F U e . We conclude that irreducible. Let U e (U e f F0 ⊗F V = (F0 ⊗F U ) ⊗F0 (F0 ⊗F W ) = IndG e 0 ⊗F0 W0 ). N e0 ⊗F W f0 is an absolutely irreducible F0 N eHence, in view of (1), Ve0 = U 0 e module, which is centralized by Z. By Mackey decomposition, identifying Γ = G/N , M G ∼ F0 ⊗F ResG Ve0τ . N (V ) = ResN (F0 ⊗F V ) = τ ∈Γ

∼ eVe for some absolutely By virtue of (3) and (2) this implies that )= e irreducible F N -module Ve satisfying F0 ⊗F Ve ∼ V = 0 , where the ramification index e = |Γ| = |G/N |. However, e = 1 since Γ is cyclic. This may be seen by observing that if θ is the Brauer character of N afforded by Ve , then the representation group Γ(θ) is a central extension of the cyclic group Γ. Hence Γ(θ) is abelian, and the Brauer character of G afforded by V is of b = θ(1). the form χ = θb ⊗ ζ for some linear character ζ of Γ(θ), where θ(1) We have the contradiction N = G. ResG N (V

e /C (W ) of G/C e (8) There is a nonabelian normal subgroup N e e(W ) G G e such that ResG e (W ) is not absolutely irreducible: N

160


e of G e which is minimal subject Otherwise consider a normal subgroup E e to the condition that its image in G/C (W ) is nonabelian. As in step e G e e (4) either (G/C e(W ), W ) is a reduced pair, its core being the image of E, G or W is obtained by (proper) tensor induction from a reduced pair with quasisimple core. Both cases are excluded by (6). (9) Conclusion: By (8), (7) and stable Clifford theory we may write W ∼ = U2 ⊗F U3 over some b e central extension G of G, where both modules Ui are absolutely irreducible b and primitive with dim F Ui < dim F W. Viewing U1 = U as an F G-module we have V ∼ = U1 ⊗F U2 ⊗F U3 . b Recall that the pair (G/C b(U1 ), U1 ) is nonreal reduced by (5). If the pairs G b (U ), U ) are nonreal reduced also for i = 2, 3, then V contains a real (G/C i b i G vector for G by Proposition 8.3c. Hence, by the choice of V , one of these modules, say U3 , has a real vector. b = U1 ⊗F U2 is an absolutely irreducible and primitive F Gb Now U b module with dimension less than d = dim F V. Thus either U contains a b or (G/C b b b real vector for G b(U ), U ) is a nonreal reduced pair. In the first G b ⊗F U3 , a case by Proposition 5.3b there is a real vector for G in V = U contradiction. By the classification of the nonreal reduced pairs the latter b = 4 and (G/C b case can hold only when F = F7 , dim F U b(U1 ), U1 ) is G 2 nonreal of type (Q8 ). However, then |U1 | = |U2 | = 7 . Since GL2 (7) is a 0 5 b b b b 50 -group, G/C b(U ) is a 5 -group too. But (G/CG b(U ), U ) is of type (2− ), G b b (2.A6 ) or (Sp4 (3)), and |G/C ¤ b(U )| is divisible by 5 in each case. G 8.5.

Nonreal Induced Pairs, Wreath Products

Let (G, V ) be a nonreal induced pair, say induced from the nonreal reduced pair (H, W ). Suppose V = W1 ⊕ · · · ⊕ Wn is an imprimitivity decomposition, where the G-conjugate reduced pairs (NG (Wi )/CG (Wi ), Wi ) are isomorphic to (H, W ). Assume that n ≥ 2. Let E be the core of (H, W ), and let Ei ∼ = E be the core of Hi = NG (Wi )/CG (Wi ). Fix an isomorphism (H1 , W1 ) ∼ = (H, W ) of reduced pairs. Inspection of Theorems 6.1, 7.1 yields that distinct normal subgroups of H


161

are not isomorphic (and in most cases even determined by their order). Let Tn N = i=1 NG (Wi ), and suppose that S = G/N acts (transitively) on the set Ω = {1, · · ·, n} like on the {Wi }. For each i ∈ Ω let Ti = CN (Wi ) and Ni = N/Ti . Identifying each Ni with N CG (Wi )/CG (Wi ) we have normal subgroups of the Hi , which are G-conjugate, hence have a common image N0 ∼ = Ni in H = H0 . Since G is faithful on V , by Lemma 8.2a we may identify G with a subgroup of Hwr S. Our objective is to show that G is not far from being T the full wreath product. Define Ri = j6=i Tj for each i ∈ Ω, which are G-conjugate normal subgroups of N (and which would agree with a direct factor Hi in the base group B = H (n) = H1 × · · · × Hn of the wreath product). We have Ri ∩ Ti = 1, and R = R1 × · · · × Rn is a normal subgroup of G. Let R0 ∼ = Ri ∼ = RCG (Wi )/CG (Wi ) be the common image of the Ri and of R in H = H0 . Observe that R0 , N0 are normal subgroups of H. Lemma 8.5a. N0 has no regular orbit on W . Hence N0 is a nonabelian normal (characteristic) subgroup of H and so N0 ⊇ E. Proof. Assume that, for some i ∈ Ω, there is wi ∈ Wi such that CNi (wi ) = 1. Since G is transitive on the {Wi }, this is true then for all i ∈ Ω. Then v = T T P i∈Ω CN (Wi ) = i∈Ω CN (wi ) = i wi is a regular vector for N as CN (v) = 1. But now by Proposition 5.3b (ii) there is a real vector in V for G, which contradicts the (implicit) hypothesis in the lemma. The remainder follows from (3.4b) and the definition of a reduced pair. ¤ For a subgroup Y of H we denote by Y S the “stabilizer residual” of Y , the intersection of all normal subgroups T of Y for which Y /T is isomorphic to a subgroup of CH (w) for some w ∈ W ] . We have to examine these residuals only when (H, W ) is of extraspecial type. Lemma 8.5b. Let (H, W ) be of extraspecial type. Suppose Y is a subgroup of H which has no regular orbit on W . Then Y S 6⊆ Z(H) when p 6= 3, 5, and Y ⊃ E for p = 5. If Y is assumed to be subnormal in H, then Y ⊇ E. Further, then Y is irreducible on U = E/Z(E) except when p = 3 or 5, or when p = 7 and (H, W ) is of type (33+ ) with Y = Ehji or EhjiZ(H) for some involution j inverting the elements of U .

162


Proof. Clearly Y is nonabelian. Let Z = Z(H). From Theorem 6.1 we know that Z ∼ and H = X = E : Sp2 (3) is the = F?p unless p = 7, E ∼ = 31+2 + standard holomorph of E. Otherwise we have H = X ◦ Z where X is the standard holomorph of E when 4 - p − 1 and p 6= 3, and of E ◦ Z4 otherwise, or we have p = 7, E ∼ and X/E ∼ = 21+2m = Ω− − 2m (2) for m = 1, 2. Recall also 5 that for p = 3 we have type (2− ), and X ∼ : (Z5 : Z4 ) is related to = 21+4 − the largest Bucht group. By Theorem 6.1 the point stabilizers either are abelian or (H, W ) is of type (25− ) with p = 7, or of type (33+ ) with p = 13, where SL2 (3) = Q8 : Z3 can appear. Hence either Y S ⊇ Y 0 or we are in these latter situations, in which cases Y S ⊇ Y 000 at least. Let χ denote the Brauer character of H afforded by W . We work through the various types, making use of the knowledge of the H-orbit structure on W ] given in Sec. 6.1. Type (Q8 ) : Here E = Q8 and X ∼ = 2− S4 or SL2 (3) for p = 7, and X ∼ = (Q8 ◦ Z4 ).S3 for p = 5 and p = 13. Let T = NH (P for some Sylow 3-subgroup P of H. Then H = ET , E ∩ T = Z(E) and T /Z ∼ = S3 or A3 . If y ∈ H is a noncentral element of order 3, then y is irreducible on U and so |χ(y)| = 1 by Theorem 4.4. Hence hZyi fixes 2 points in P1 (W ) if 3 | p − 1, and at most 1 otherwise. Each noncentral involution in H fixes 2 points in P1 (W ). It follows that T has a regular orbit on W for p = 13. The same holds true for p = 7, because then P = Z3 × CH (w) for some w ∈ W ] (Sec. 6.1), and the involutions in T /Z leave F w invariant. For p = 5 any such stabilizer CH (w) is cyclic of order 4, and either T is regular on W ] or has two orbits of size 12. But in the latter case H/Z = (EZ/Z) : (T /Z) has an element of order 4 whose square is in T /Z, which is impossible. Let next T be a subgroup of H such that Z(E) ⊂ E ∩ T ⊂ E. Then T contains no element of X of order 3 that is irreducible on U . It follows that T /Z has order 4 and so has a regular orbit on P1 (W ). For p = 5 argue as above, in which case T can be dihedral of order 8 (having two regular orbits). Thus Y ⊇ E. Of course E ∼ = Q8 has a regular orbit on W , and EZ has a regular orbit on W if and only if p 6= 5. Hence Y S ⊇ Y 0 6⊆ Z for p 6= 5, as desired. Finally, if Y is subnormal in H and not irreducible on U , then Y = EZ and p = 5. Type (25− ) : Here E ∼ and X = E : (Z5 : Z4 ) for p = 3, and = 21+4 − ∼ X = E.S5 or E.A5 for p = 7. The core E has 10 noncentral involutions


163

corresponding to the 5 singular points in the orthogonal space U = E/Z(E). Each such involution t has two eigenspaces on W , both of dimension 2, and so fixes exactly 2 · (p + 1) points in P1 (W ). Since t and tz for z ∈ Z(E) have the same eigenspaces, the number of points in P1 (W ) fixed by some noncentral involution in E is at most 5 · 2 · (p + 1). This number is smaller than |P1 (W )| = p3 + p2 + p + 1 for p = 7 but not for p = 3. Indeed for p = 3 there is no regular E-orbit on W . Here H is transitive on W ] , a point stabilizer being cyclic of order 8 (Sec. 6.1). Since E is normal in H, the E-orbits on W ] have equal size, but |W ] | = 80 is not divisible by |E| = 25 . Each proper subgroup of E has a regular orbit on W , however. Suppose Y is subnormal in H. Then (Y ∩ E)/Z(E) cannot be a proper subgroup of U , because the normal closure of Y in Y E then were a proper subgroup of Y E and so Y centralizes a nontrivial quotient group of U . But this implies that Y is a 50 -group, and that Y ⊂ EZ by the structure of H. This in turn forces that p = 3 and Y = E. For p = 7 we obtain that Y = X or X 0 and so Y is irreducible on U . Let p = 7 in what follows, and assume that Y is not subnormal in H. In the proof for Proposition 6.6b, part (ii), we have seen that there is, up to conjugacy, a unique subgroup T ∼ = 2.S5 of X such that X = T E and X T ∩ E = Z(E), and ResT (χ) is the character obtained by fusing χ6 and χ7 in the Atlas notation (p. 2). Let y ∈ T be an element of order 3. Then (2) (2) χ(y) = −2 and yW = [z1 , z2 ] where z1 6= z2 are the primitive 3rd roots of unity. Hence y fixes 2 · 8 = 16 points in P1 (W ). The group EZhyi has a regular orbit on W , because EZhyi/Z contains |E : CE (y)| = |χ(y)|2 = 4 (conjugate) subgroups of order 3 and 80 + 4 · 16 < |P1 (W )|. Each noncentral involution x ∈ T (χ(x) = 0) fixes at most 2 · 8 = 16 points in P1 (W ). There are 16 subgroups isomorphic to Zhxi/Z in EZhxi/Z complementary to EZ/Z. Hence EZhxi has a regular orbit on W since 80 + 16 · 16 < 400 = |P1 (W )|. The subgroup T0 ∼ = 2.A5 of T has a regular orbit on W but T ◦ Z does not (see the remark to Theorem 7.2a). However this does not matter since Y /Y S is solvable. It remains to examine the situation that Y ∩ E ⊃ Z(E). We know that Y ZE/ZE is not trivial and not of order 2 or 3. Suppose first that Y maps onto SL2 (3) or onto Q8 . Then Y ZE/ZE is isomorphic to A4 resp. to Z2 × Z2 . In both cases Y centralizes a singular point in U but no nonsingular one [Atlas, p. 2]. By definition Y S ⊆ Y ∩ E and [Y ∩ E, Y ] ⊆ Y S .

164


Thus from Y S ⊆ Z it follows that (Y ∩ E)/Z(E) does not contain nonsingular points. In other words, Y ∩ E is elementary of order 4. But then a simple counting argument shows that Y has a regular orbit on W . Hence we are left with the case that Y S ⊇ Y 0 . If |Y | is divisible by 5, then Y is irreducible on U . Then even Y 0 ⊇ E. Then Y S ⊆ Z implies that Y Z/Z is abelian and that Y is a {2, 3}-group. Moreover, Y ZE/ZE must be isomorphic either to Z2 × Z2 or to S2 × A3 . The intersection of the centralizers in S5 ∼ = O− 4 (2) of a singular and a nonsingular point of U has order at most 2 [Atlas]. We conclude that in the former case Y ∩ E either is elementary abelian of order 4, which is handled as above, or (Y ∩ E)/Z(E) does not contain singular points. In the S2 × A3 case Y also centralizes no nonsingular point. Thus Y ∩ E is cyclic (of order 4) or isomorphic to Q8 in these cases. Again a simple counting argument shows that Y has a regular orbit on W , in contrast to our hypothesis. ∼ Type (33+ ) : Here E ∼ = 31+2 + , X = E : Sp2 (3) and p = 7 or 13. Each noncentral element of E has 3 eigenspaces on W to different eigenvalues. Hence every nontrivial subgroup of E/Z(E) fixes just 3 points in P1 (W ). There are 3 + 1 = 4 such subgroups. Thus the number of points in P1 (W ) which are fixed by some nontrivial element of E/Z(E) is at most 4 · 3 = 12, which is less than |P1 (W | = (p3 − 1)/(p − 1) for p = 7 and for p = 13. So EZ has a regular vector in W . If x is an element of order 3 in X outside E, then by Theorem 4.4 either CE (x) = Z(E) and χ(x) = 0, or CE (y)/Z(E) = CU (x) has order |χ(x)|2 = 3. In the former case x fixes just three points in P1 (W ), otherwise at most 1 + (p + 1) points, among them the eigenspaces of the noncentral elements of E contained in CE (x). If j is an involution in X, then j inverts the elements of U and |χ(j)| = 1 by Theorem 4.4. By Theorem 4.5a, or by inspection of the character table of Sp2 (3), χ(j) = −1. Since Sp2 (3) has four elements of order 3 and a unique involution, and since 4 · (p + 2) + 1 · (p + 2) < |P1 (W )| (for p = 7 and p = 13), every complement to EZ/Z in H/Z has a regular vector in W . If T is a subgroup of H with Z(E) ⊂ E ∩ T ⊂ E, then T /Z cannot contain an element of order 4. So consider T = ZE0 hx, ji where x, j are as above and E0 is a subgroup of E of order 32 normalized by x. Either x centralizes E0 , and then E0 Z/Z and Zhxi/Z together fix at most p + 2 points in P1 (W ), or these fix 2 · 3 ≤ p + 2 points. There are four subgroups in T /Z of order 3, and there are 3 subgroups in T /Z of order 2. Thus at


165

most 3 · (p + 2) + 3 · (p + 2) points in P1 (W ) are fixed by some nontrivial element of T /Z. It follows that T has a regular orbit on W . Consequently Y ⊃ E. Assume that Y is subnormal in H but not irreducible on U . Then Y does not contain an element of order 4, and no element of order 3 outside E, whence |Y Z/EZ| = 2. There are |U | = 9 involutions j in Y , hence at most 9(p + 2) elements in P1 (W ) are fixed by any of them. Since the number of points in P1 (W ) fixed by nontrivial subgroups of EZ/Z is at most 12, and since 12 + 9 · (p + 2) < |P1 (W )| for p = 13, we have p = 7. In the p = 7 case the result is as stated. In general Y ⊃ E contains an involution j inverting the elements of U , or an element x ∈ X of order 3 outside E. In each case Y /Y ∩ Z is nonabelian, and Y S 6⊆ Z as the nonabelian point stabilizers are isomorphic to SL2 (3). ¤ Remark . Let (H, W ) be the unique, up to isomorphism, nonreal reduced pair in characteristic p = 5. Then H is a 5-complement in GL2 (5); the pair (H, W ) is of type (Q8 ) and will turn out to be exceptional (Chapter 10). We have seen above that if Y is a subgroup of H having no regular orbit on W , then Y properly contains the core E ∼ = Q8 of H. Indeed either Y is one (of three) normal subgroups of H containing O2 (H) ∼ = Q8 ◦ Z4 , or Y =P ∼ = Z4 wr Z2 is a Sylow 2-subgroup of H. Here P 0 is cyclic of order 4 and subnormal but not normal in H (see Lemma 10.4a below). Theorem 8.5c (Riese–Schmid). Let (G, V ) be nonreal induced. Keeping the notation introduced above we have R0 ⊇ E, except possibly when p = 3 or 5. In the exceptional cases at least R0 ⊇ Z(E) (and N0 ⊇ E). Proof. Since there is no real vector in V for G, N has no regular orbit on V by Proposition 5.3b. From Lemma 8.5a we know that N0 has no regular orbit on W , and N0 ⊇ E. Our argumentation will be different depending on whether the core E of (H, W ) is quasisimple or of extraspecial type. Case 1: E is quasisimple Assume the assertion is false. Then R0 6⊇ E. Since R0 is normal in H, by definition of reduced pairs this implies that R0 is abelian and R0 ⊆ Z(H). So for each i ∈ Ω, using that G is transitive on Ω and identifying ¯ i ⊆ Z(Ni ) where Ni = N/Ti with a subgroup of Hi , we have Ni ⊇ Ei and R ¯ Ri = Ri Ti /Ti . Also, for i 6= j in Ω, Tij = Ti Tj is a normal subgroup of N

166


with Tij /Ti ∼ = Tij /Tj and Tij /(Ti ∩ Tj ) ∼ = Tij /Ti × Tij /Tj . Either Tij /(Ti ∩ Tj ) is abelian and hence central in N/(Ti ∩ Tj ), or Ei ⊆ Tij /Ti and Ej ⊆ Tij /Tj . Define the subgroup Zi of N by Zi /Ti = Z(N/Ti ) (i ∈ Ω), and define a binary relation on Ω by letting i ∼ j if Zi = Zj . This is a G-invariant equivalence relation on Ω. We assert that each equivalence class J has at least two elements. In order to prove this we use that E is quasisimple. T T Note that i Ti = 1 and that i Zi = Z(N ). There exists a quasisimple subnormal subgroup Y of N , a component of N . Either Y Ti /Ti ∼ = Ei or Y ⊆ Ti . In the former case Y ∼ = Ei . (Use ∼ that if the multiplier M(E) 6= 1, then E = 2.A6 and M(E) = Z3 , but Y T maps into Ni0 = Ei and Z(Y ) ⊆ i Zei where Zei /Ti corresponds to Z(Ei ).) Let J = JY be the set of those j ∈ Ω for which Y is mapped isomorphically T onto Ej . Then J 6= ∅, and if J = {j} then Y ⊆ Rj = i6=j Ti were ¯ j ⊆ Z(Hj )). Hence |J| > 1. Let i 6= j be in J. Assume abelian (as R i 6∼ j. Then Y = Y 0 maps isomorphically onto a subnormal subgroup of Tij /(Ti ∩Tj ) which in turn maps onto the normal subgroup Ei of Tij /Ti and on the normal subgroup Ej of Tij /Tj . This is impossible since Tij /(Ti ∩ Tj ) has only 2 components [Aschbacher, 1986, (31.7)]. Alternately use that in a direct decomposition of perfect groups the direct factors are uniquely determined. Thus J = JY is an equivalence class with respect to ∼, hence a G-block on Ω, with |J| ≥ 2. For if g ∈ G is such that Y g = Y , then g stabilizes J, and if Y g 6= Y , then [Y, Y g ] = 1 and J g = JY g intersects J trivially. Pick i 6= j in J, and write ZJ = Zi = Zj . We assert that there are nonzero vectors wi ∈ Wi and wj ∈ Wj such that CN (wi ) ∩ CN (wj ) ⊆ ZJ . By Theorem 7.1 we may choose wi such that CN (wi )/Ti is cyclic, except when (H, W ) is of type (Sp4 (3)) with p = 7, 13. In the cyclic case the group C = CN (wi )ZJ /Tj is abelian, and we pick wj in a regular C-orbit on Wj . In the exceptional cases we treat the (worst) situation that Ni = N/Ti is isomorphic to Sp4 (3) × Z(p−1)/2 so that ZJ /Ti ∼ = F∗p . By Theorem 7.1 we may pick wi ∈ Wi such that CN (wi )/Ti is isomorphic to SL2 (3) × Z3


167

(p = 7) resp. to Z3 wr S2 (p = 13). By Corollary 7.2c, CN (wi )ZJ /ZJ has a regular orbit on P1 (Wi ), and this depends only on the isomorphism type of the point stabilizer. It follows that CN (wi )ZJ /Tj has a regular orbit on Wj likewise. Hence the assertion. P Thus there is vJ = j∈J wj consisting of nonzero vectors wj ∈ Wj T such that CN (vJ ) = j∈J CN (wj ) ⊆ ZJ . Let v be the sum of the distinct G-conjugates of vJ . Then \ CN (v) ⊆ (ZJ )g = Z(N ). g∈G

So CN (v) acts on each Wi as a group of scalar multiplications fixing a nonzero vector. This implies that CN (v) = 1, a contradiction. Case 2: E is extraspecial Here we argue as follows. Let the base group B = H1 × · · · × Hn of the wreath product Hwr S act on V = W1 ⊕ · · · ⊕ Wn diagonally (as usual). Suppose N is any subgroup of B which has no regular orbit on V and for which the image Ni of N in Hi is a subnormal subgroup of Hi , for all i ∈ Ω. T ¯ i = Ri Ti /Ti (as usual). We assert Let Ti = CN (Wi ), Ri = j6=i Tj and R ¯ ¯ i ⊇ Ei for some that Ni ⊇ Ei and Ri ⊇ Z(Ei ) for some i ∈ Ω, and even R i if p 6= 3 and p 6= 5. We argue by induction on |N |. So we ignore, at first, the group G and its transitive action on Ω. Only in a concluding step this will be used (yielding then the assertion for all ¯i i ∈ Ω, hence for the common images N0 of the Ni and R0 of the Ri ∼ =R in H = H0 ). ¯ i 6⊇ Ei for all i ∈ Ω, and Assume that this assertion is false (so that R ¯ even Ri 6⊇ Z(Ei ) when p = 3 or 5). If for each i, Ni has a regular orbit on Wi , so does N on V . (Argue as for Lemma 8.5a.) So we may assume that N1 , say, does not have a regular vector in W1 . Then N1 ⊇ E1 by Lemma 8.5b. We have R1 ∩ T1 = 1, and T1 contains R2 , · · ·, Rn . Similarly e = R1 · T1 fulfills T1 acts faithfully on Ve = W2 ⊕ · · · ⊕ Wn . The group N ∼ e all requirements. Since N1 contains E1 but N /T1 = R1 does not, we have e | < |N |. Since C (L Wj ) ⊆ Ri for all i, by induction there must be |N j6=i e N e . We may write uniquely v = w1 + ve with a regular vector v ∈ V for N Pn w1 ∈ W1 and ve = i=2 wi for wi ∈ Wi . Here without loss we may assume that all wi 6= 0. Let C1 = CN (e v) =

n \ i=2

CN (wi ).

168


Then C1 contains R1 as a normal subgroup. Clearly C1 ∩T1 = CN (v)∩T1 = 1. So C¯1 = C1 T1 /T1 is a subgroup H1 which is isomorphic to C1 and ¯ 1 . If there exists w1 in W1 which is regular for C1 , then w1 + ve contains R is a regular vector in V for N , a contradiction. Consequently C1 has no regular orbit on W1 . Consider the action of C1 on Wi for i = 2, · · ·, n. Let C¯i = C1 Ti /Ti . Then C¯i ⊆ CH (wi ) for i ≥ 2. These point stabilizers are well understood i

by Theorem 6.1. It follows that C1S ⊆ R1 . Let first p 6= 3, 5. Suppose that N1 is irreducible on U1 = E1 /Z(E1 ). ¯ 1 ⊆ Z(N1 ) ⊆ Z(H1 ). But then C¯ S ⊆ Z(H1 ), which contradicts Then R 1 Lemma 8.5b. Thus N1 is not irreducible on U1 (but still N1 ⊇ E1 ). From Lemma 8.5b it follows that then (H, W ) ∼ = (Hi , Wi ) must be of type (33+ ) and p = 7. Moreover, then N1 = E1 · hji or E1 hjiZ(H1 ) for some involution j of H1 inverting the elements of U1 . Since C1 has no regular orbit on W1 , C¯1S 6⊆ Z(H1 ) by Lemma 8.5b. Since C¯1S is contained in all (normal) subgroups of C¯1 with index 3 (as Z6 is a point stabilizer) and E1 is a 3-group, this implies that C¯1 = E1 hji or E1 hjiZ(H1 ). Since j inverts the ¯ 1 6⊇ E1 , C1 /R1 has an S3 quotient group. However, by elements of U1 and R the structure of the point stabilizers C1 /C1S does not have an S3 quotient group. This is the desired contradiction. ∼ 21+4 ) or p = 5 (E = ∼ Q8 ). Then all point Let finally p = 3 (E = −

stabilizers of nonzero vectors are cyclic 2-groups by Theorem 6.1. By as¯ 1 6⊇ Z(E1 ). Since R ¯ 1 is normal in N1 ⊇ E1 , this forces that sumption R S R1 = 1 = C1 . Hence C1 is an abelian 2-group. But then C1 has a regular orbit on W1 , a final contradiction. This completes the proof. ¤ In the nonreal induced (imprimitive) situation k(GV ) ≤ k(N V ) · k(S) by (1.7b). Here S = G/N is a transitive permutation group of degree n ≥ 2. The class number k(S) will be investigated in the next chapter. When p 6= 3, 5 by Theorem 8.5c the group N contains a normal subgroup of G which is the direct product of n copies of the core E of (H, W ) permuted transitively by S. So G is indeed not far from being the full wreath product Hwr S. Observe that if G = Hwr S, then GV = (HW )wr S. Let us compute k(Xwr S) for an arbitrary finite group X and an arbitrary transitive permutation group S of degree n ≥ 2. Let k = k(X), and let Σn,s be the set of partitions of the n-set Ω = {1, · · ·, n} into s parts, so that σn,s = |Σn,s | is the Stirling number of the second kind to n


169

and s. Counting the ordered n-sets of a k-set with repetition one gets the combinatorial identity n X

k(k − 1) · · · (k − s + 1)σn,s = k n .

s=1

This identity is valid even when k < n (where the summands for s > k vanish). For a partition σ the stabilizer Sσ is the set of all elements of S belonging to the Young subgroup of Sn determined by σ. Proposition 8.5d. For k = k(X) we have Pn P k(Xwr S) = s=1 k(k − 1) · · · (k − s + 1) σ∈Σn,s k(Sσ )/|S : Sσ |. If S = Zn is generated by an n-cycle, then k(Xwr S)=(k n − k)/n + kn if n is a prime, and k(Xwr S) ≤k n − k + kn in general. Proof. Every irreducible character χ of the base group X (n) of G = Xwr S is of the form χ = χ1 ⊗···⊗χn for unique irreducible characters χi ∈ Irr(X). This determines a partition σ(χ) of Ω consisting of those subsets on which the components χi agree and which are maximal with this property. The inertia group IS (χ) = Sσ(χ) . One knows that χ can be extended to the inertia group Xwr Sσ(χ) of χ in Xwr S [James–Kerber, 1981, p. 154]. If σ ∈ Σn,s is a partition of Ω into s parts, there are k(k − 1) · · · (k − s + 1) irreducible characters χ of X (n) with σ(χ) = σ if k ≥ s, and none otherwise. Hence k(G) is as stated. Let S = Zn = hxi. If n is a prime, then Sσ = 1 unless σ = Ω. Hence k(G) = k|S| + (k n − k)/|S| by virtue of the above combinatorial identity. In general the estimate follows, in a similar manner, once we have shown that X k(Ss )/|S : Sσ | ≤ σn,s σ∈Σn,s

for each s ≥ 2. Suppose σ ∈ Σn,s is such that Sσ 6= 1. Let ∆ be a set in σ of n smallest cardinality. Then |∆| = t > 1 is a divisor of n and Sσ = hx t i (and each set belonging to σ has cardinality divisible by t). Pick any α ∈ Ω r ∆, say α ∈ Γ ∈ σ, and replace ∆ by ∆ ∪ {α}, Γ by Γ r {α} and leave the other sets in σ unchanged. We obtain σα ∈ Σn,s with Sσα = 1. Noting that t ≤ ns and that at most n−1 n σn,s partitions in Σn,s have nontrivial s σn,s σn,s stabilizer in S, the result follows since obviously sn2 n− n + n ≤ σn,s . ¤ s

Chapter 9

Class Numbers of Permutation Groups Let Ω = {1, · · ·, n} for some integer n ≥ 2. Permutation groups of degree n are regarded as acting on Ω. For each such group S one has the upper bound k(S) ≤ 2n−1 , due to [Liebeck and Pyber, 1997]. We shall give a sketch of proof for this theorem, and present a slight improvement. 9.1. The Partition Function Since two elements of the symmetric group Sn are conjugate if and only if they have the same cycle type, k(Sn ) = p(n) where p(n) is the partition function on positive integers. There is no simple formula for p(n). Hardy √ and Ramanujan showed that asymptotically p(n) ∼ yields the estimate (9.1a)

p(n) < √

π 6(n−1)

· eπ

2n/3 eπ √ , 4n 3

which readily

√

2n/3

for n ≥ 2 (cf. [Erdös, 1942] for√an elementary approach). We get the upper bound p(n) < 0.2735 · 23.701 n , which is fairly good for n ≥ 23. With elementary means one can establish the following. Proposition 9.1b. For n ≥ 5 we have k(An ) < k(Sn ) ≤ 2n−2 . Proof. The assertion is verified for 5 ≤ n < 10 by inspection [Atlas]. Let n ≥ 10. Let kd be the number of conjugacy classes of elements in Sn with shortest cycle length d, so conjugate to (12···d). Then either d = n (kn = 1) or d ≤ b n2 c. Consider the elementwise stabilizer S(d) ∼ = Sn−d of {1, 2, · · ·, d} for d 6= n in Sn . Evidently kd ≤ k(S(d) ). By induction k(S(d) ) < 2n−d−2 for n 6= d since then n − d ≥ 5. Thus n

k(Sn ) < 1 + 2n−3 + 2n−4 + · · · + 2d 2 e−2 , which is certainly less than 2n−2 . Let x ∈ An . If xSn = xAn , then associate to x the partition σ defined by x ∈ Sn . If xSn 6= xAn , then σ is of the form (a1 , · · ·, as ) with a1 > a2 > 170

Class Numbers of Permutation Groups

171

··· > as and all aj being odd. In this case xSn consists of two An -classes, and we associate to these the different partitions σ and σ 0 = (a1 −1, a2 , ····, as , 1). Hence k(An ) ≤ k(Sn ). Since the partition (n − 3, 3) is not associated to a conjugacy class of An in this manner, we even have k(An ) < k(Sn ). ¤ One knows that roughly k(An ) is half of k(Sn ) (Erdös). 9.2. Preparatory Results We need some preparations. For a simple group L we denote by P (L) the smallest degree of a faithful permutation representation of L, that is, the index of a proper subgroup of L of largest order. Thus P (L) ≥ R0 (L) + 1 and P (L) ≥ Rp (L) + 1 for each prime p. Lemma 9.2a. Suppose L is a finite simple group of Lie type of (untwisted) rank ` = rk(L) over Fr , where r = q t for some prime q. (i) If ` = 1, then L = L2 (r) = PSL2 (r) and P (L) ≥ r + 1 unless r ≤ 11. (ii) If L =2 B2 (r) = Sz(r) is a Suzuki group, then P (L) = r2 + 1, and if L =2 G2 (r) is a Ree group, then P (L) = r3 + 1. (iii) In general P (L) ≥ (r`+1 − 1)/(r − 1) except when L = L2 (7), L2 (11), G2 (4), 3 D4 (2), 2 F4 (2)0 , 2 B2 (r) or 2 G2 (r). Statement (i) is a classical result of Galois (and Dickson) [Huppert, 1967, II.8.28]. The result on the Suzuki groups in (ii), where r = 2t with t ≥ 3 odd, is due to [Suzuki, 1962], and the result on the Ree groups, where r = 3t with t ≥ 3 odd, can be found in [Ward, 1966]. (The Ree groups 2 F4 (2t ), t ≥ 3 odd, are not exceptional here.) The general lower bounds for P (L) can be found in Table 5.2A for L classical, and in Table 5.3A for L exceptional in [Kleidman–Liebeck, 1990]. In the cases L = G2 (4), 3 D4 (2) and for the Tits group 2 F4 (2)0 one can take P (L) from the [Atlas], namely P (L) = 416, 819 and 1.600 respectively. Lemma 9.2b. Let L, ` = rk(L) and r = q t be as before. Then |Out(L)| ≤ 2(` + 1)t exept when L ∼ = PΩ± 8 (r), in which case |Out(L)| ≤ 24t. This is immediate from Steinberg’s description of the automorphisms of the groups of Lie type in Chapter 10 of [Steinberg, 1967]. For L = L2 (r) we have |Out(L)| = 2t if r is odd and |Out(L)| = t otherwise.

172


Lemma 9.2c. Let L, ` = rk(L) and r = q t be as before. (i) k(L2 (r)) ≤ r + 1, k(Sz(r)) = r + 3 and k(2 G2 (r)) = r + 8. (ii) In general k(L) ≤ (6r)` . [Schur, 1911] has already computed the character table of L2 (r), [Suzuki, 1962] for the Suzuki groups (see also [Huppert–Blackburn, 1982, XI.5.10]), and [Ward, 1966] for the Ree groups. Hence the statements in (i). Statement (ii) is due to [Liebeck and Pyber, 1997]. It is based on general results on conjugacy classes in algebraic groups [Springer–Steinberg, 1970]. For a proof of the following fundamental result we refer to [Dixon– Mortimer, 1996]. Theorem 9.2d (O’Nan–Scott). Suppose S is a primitive permutation group of degree n. Then the socle of S has the form L(m) for some simple group L, and one of the following holds: • Affine type: |L| = q for some prime q, S is a subgroup of the affine group AGLm (q) containing the translations, and a point stabilizer in S is an irreducible subgroup of GLm (q). • Diagonal type: L is nonabelian (simple), n = |L|m−1 with m ≥ 2 and S is a subgroup of a wreath product with the diagonal action. Also, S/L(m) is isomorphic to a subgroup of Out(L) × Sm , and a point stabilizer has a primitive action of degree m. • Product type: L is nonabelian, n = bd for integers b, d greater than 1 and S is a subgroup of T wr Sd , in product action, for some primitive subgroup T of Sb . • Almost simple type: S is almost simple (m = 1 and S ⊆ Aut(L)). 9.3. The Liebeck–Pyber Theorem Theorem 9.3 (Liebeck–Pyber). If S is a permutation group of degree n, then k(S) ≤ 2n−1 . Proof. We argue by induction on n. Suppose S is not transitive, and let Ω1 be an S-orbit and Ω2 = Ω r Ω1 . Then N = CS (Ω1 ) is faithfully represented on Ω2 as a permutation group of degree n2 = |Ω2 | < n, hence k(N ) ≤ 2n2 −1 by induction. S/N is isomorphic to a subgroup of Sn1 where n1 = |Ω1 |, whence k(S/N ) ≤ 2n1 −1 . By (1.7b) k(S) ≤ k(N ) · k(S/N ) ≤ 2n1 −1+n2 −1 = 2n−2 .


173

Let T = S1 be a point stabilizer, and assume T is not maximal in S. Choose a proper intermediate group T ⊂ H ⊂ S, and let C = CoreG (H). Let |S : H| = b, |H : T | = d, and let b0 = |S : CT |, d0 = |CT : H|. By induction k(S/C) ≤ 2b−1 as S/C is isomorphic to a subgroup of Sb . Also, C has b0 orbits on Ω = {1, · · ·, n} of size d0 , and the preceding argument yields k(C) ≤ (2d0 −1 )b0 . By (1.7b) k(S) ≤ 2(d0 −1)b0 · 2b−1 = 2n+b−b0 −1 ≤ 2n−1 . because b0 d0 = n and b ≤ b0 . Hence we are reduced to the case that S is primitive. Let first S ⊆ AGLm (p) be of affine type (n = pm ), and let M = CS (M ) be a minimal (regular) normal subgroup of S. By Theorem 2.3c the principal block is the unique p-block of S. Therefore by the Brauer–Feit Theorem 2.4, 1 k(S) ≤ 1 + |P |2 4 where P is a Sylow p-subgroup of S. Since P/M is isomorphic to a subgroup of a Sylow p-subgroup of GLm (p), which has order pm(m−1)/2 , we have k(S) ≤ 1 + 41 pm(m+1) . We have to examine the cases where 1 + 41 pm(m+1) > m m 2p −1 , that is, where pm(m+1) ≥ 2p +1 . This happens only when p = 2 and m ≤ 4. Here the cases m = 1, 2, 3 are easily treated. Note that every subgroup of GL3 (2) has at most 7 conjugacy classes. In the m = 4 case consider the stabilizer C = CS (w) for some w ∈ M ] , so that C/M is isomorphic to a subgroup of a parabolic subgroup R : GL3 (2) of GL4 (2), (3) with unipotent radical R ∼ = Z2 , and |S : C| ≤ 15. Using (1.7b) we conclude that k(S/M ) ≤ 15 · 7 · 8 < 210 and k(S) ≤ 214 , as desired. Let next S be of diagonal type. So the socle M of S is the direct product of m > 1 copies of a nonabelian simple group L, and n = |L|m−1 . Further S/M is isomoprphic to a subgroup of Out(L) × Sm . Clearly k(L) ≤ |L|/2 and |L| ≥ 60. From the classification of the finite simple groups and Lemma 9.2b one has the very crude estimate |Out(L)| ≤ |L|. (If L is an alternating or sporadic simple group, then |Out(L)| ≤ 2, except when L = A6 where |Out(L)| = 4.) Consequently k(S) ≤ k(L)m |Out(L)| · k(Sm ) ≤ |L|m+1 · 2−m · 2m−1 ≤ |L|m+1 , which gives the result.

174


Let S be of product type. So n = bd for integers b, d greater than 1, and S is a subgroup of T wr Sd for some primitive subgroup T of Sb . By induction we know that every subgroup of T has at most 2b−1 conjugacy classes. Hence every subgroup of the base group N = T (d) has at most 2(b−1)d conjugacy classes. Since S/S ∩ N is isomorphic to a subgroup of Sb , we have k(S/S ∩ N ) ≤ 2d−1 . By (1.7b) k(S) ≤ 2(b−1)d · 2d−1 ≤ 2bd−1 . Of course bd ≤ bd . So let S be almost simple, say L ⊆ S ⊆ Aut(L) for some simple nonabelian group L. If L = An for n 6= 6, then Aut(L) = Sn and the result follows (see above). If L = A6 or if L is a sporadic simple group, the desired conclusion follows by inspection of the Atlas. Use that n is larger than the degree of each faithful character. So let L be a group of Lie type over Fr , and L ∼ 6 An . Let r = q t = for some prime q, and let ` = rk(L). Clearly n ≥ P (L). We use Lemmas 9.2a, 9.2b and 9.2c, and we use (1.7b). The cases where L = L2 (r) or L is a Suzuki or Ree group are easily treated using the information given above. The groups G2 (4), 3 D4 (2) and 2 F4 (2)0 are treated with the help of the Atlas. In the remaining cases we have k(S) ≤ (6r)` |Out(L)| ≤ 2(r as desired. 9.4.

`+1

−1)/(r−1)−1

≤ 2n−1 , ¤

Improvements

Recently [Maróti, 2005] has shown√that, for n ≥ 3, the Liebeck–Pyber bound can be improved to k(S) ≤ 3 n−1 . When S is solvable or |S| not divisible by 3, 5 or 7, this stronger bound had been established previously by [Kovács–Robinson, 1993] and [Riese–Schmid, 2003], and this suffices for our purposes. Lemma 9.4a. Let p be any of the primes 3, 5 or 7. If S is a primitive n permutation group of order prime to p and of degree n ≥ 7, then k(S) ≤ 2 2 . Proof. Of course we invoke the O’Nan–Scott Theorem 9.2d. Suppose first that S is of affine type. Then S has a minimal normal q-subgroup M for some prime q 6= p, with CS (M ) = M and |M | = n = q m for some m. Also S = M : H where H is an irreducible p0 -subgroup of GLm (q). Arguing as


175

before we have to examine the cases where 1 + 41 q m(m+1) > where m q m(m+1) ≥ 2(q +4)/2 .

√ qm 2 , that is,

This implies that q = 7, m = 1 or q = 3, m = 2, 3 or q = 2, m = 3, 4, 5 (noting that n = q m ≥ 7). If (q, m) = (7, 1), then p = 3 or 5 and k(S) ≤ n = 7. If (q, m) = (3, 2), then p = 5 or 7 and H may be any irreducible subgroup of GL2 (3). If 3 - |H|, then k(S) ≤ 1 + 41 34 by (2.4), hence k(S) ≤ 21 ≤ 29/2 . Otherwise H = GL2 (3) or SL2 (3), in which cases we get k(S) = 11, 10, respectively. Let (q, m) = (3, 3). Then again H may be any irreducible subgroup of GL3 (3). If H = GL3 (3), then k(S) = k(GL3 (3)) + k(AGL2 (3)) = 24 + 11 = 35 by (1.10b), because the stabilizer in H of any nontrivial linear character λ of M is isomorphic to AGL2 (3) and λ extends to IS (λ) [Atlas, p. 13]. For H = SL3 (3) we get k(S) = 12 + 11 = 23. In the remaining cases H is 3 solvable of order dividing 72, and |S| ≤ 23 /2 . Let (q, m) = (2, 3). If H = GL3 (2), then p = 5 and k(S) = k(GL3 (2))+ k(S4 ) = 6 + 5 = 11 by (1.10b) since the stabilizer IH (λ) ∼ = S4 for any nontrivial linear character λ of M . Otherwise H is cyclic of order 7 or is a Frobenius group of order 21, and we get k(S) ≤ 8. Let (q, m) = (2, 4). Then we must have p = 7 and, by inspection of the possible irreducible 70 -subgroups of GL4 (2), k(S) ≤ k(42 : GL2 (4)) ≤ m 2 · (15 + 4) ≤ 2q /2 . For (q, m) = (2, 5), H must be cyclic of order 31 or a Frobenius group of order 31 · 5 (when p 6= 5), and the result follows. Suppose next that S is of product type. Here n = bd for certain integers b, d greater than 1 and S is a subgroup of T wr Sd , T being a primitive subgroup of Sb . Let N denote the base group of T wr Sd . By Theorem 9.3 each subgroup of T has at most 2b−1 conjugacy classes. Thus k(S ∩ N ) ≤ 2(b−1)d . Since S/S ∩ N is isomorphic to a subgroup of Sd we have k(S/S ∩ N ) ≤ 2d−1 , again by Theorem 9.3. The result follows if bd − 1 ≤ bd /2. Since n = bd ≥ 7, this is true unless b = 2, d = 3 or b = 3, d = 2. The primitive permutation p0 -groups of degree 8 or 9 are solvable, hence are of affine type and already treated. Suppose that S is of diagonal type. Here the socle M of S is the direct power of m > 1 copies of some simple p0 -group L, n = |L|m−1 and S/M is isomorphic to a subgroup of Out(L) × Sm . Hence k(S) ≤ k(L)m · |Out(L)| · 2m−1 ≤ |L|m+1 ≤

√ |L|m−1 2 ,

176


because k(L) ≤ |L|/2, |Out(L)| ≤ |L| and |L| ≥ 60. So we may assume that S is almost simple, say L ⊆ S ⊆ Aut(L) for some simple p0 -group L. The alternating groups are ruled out by assumption. Also, the case p = 3 is easy since then L = Sz(q) is a Suzuki group, and Lemmas 9.2a, 9.2c apply. The only sporadic groups meeting the requirements are M11 , M12 and J3 for p = 7, for which the result holds [Atlas]. For groups of Lie type we argue as in the proof for Theorem 9.3. The exceptional groups in Lemma 9.2a, including the Ree groups 2 G2 (r), and the groups of rank 1 (L = L2 (r)) are treated as before. Letting L be defined over of Fr and ` = rk(L), as usual, we therefore have ` ≥ 2 and n ≥ P (L) ≥ (r`+1 − 1)/(r − 1), and it suffices to verify the inequality 1

(6r)` |Out(L)| ≤ 2 2 (r

`+1

−1)/(r−1)

.

Using the bound for |Out(L)| given in Lemma 9.2b this inequality holds unless ` = 2, r ≤ 4 or ` = 3, 4 and r = 2. The corresponding p0 -groups L are all in the Atlas, namely the 50 -groups A2 (3), 2 A2 (3) ∼ = G2 (2)0 , G2 (3) − 3 0 2 ∼ and D4 (2), and the 7 -groups A3 (2) = Ω6 (2), C2 (4) = PSp4 (4), 2 A2 (4), 2 A4 (2) and 2 F4 (2)0 . The desired√estimate holds in each case. (Except for √ 5 ¤ L = A2 (3) one can even replace 2 by 2.) Proposition 9.4b. Let again p be any of the primes 3, 5, 7, and let S ⊆ Sn be a p 0 -group. Suppose S has r orbits of size 2 and s orbits not of size 2. √ n−2r−s √ n−1 . In particular, k(S) ≤ 3 for n ≥ 3. Then k(S) ≤ 2r · 3 Proof. Assume the proposition is false, and let S be a counterexample with n minimal. Then S is transitive. For if S has an orbit of size 2, then √ n−2−2(r−1)−s by induction, using (1.7b). If S has no k(S) ≤ 2 · 2r−1 · 3 orbit of size 2 but is not transitive, argue as in the proof for Theorem 9.3. By inspection n ≥ 5. Suppose we have shown that S is even primitive. Then Lemma 9.4a yields that n ≤ 6. For n ≤ 4 the result follows by inspection. For n = 5 we have k(S) ≤ 7. There are just four primitive (70 -) groups of degree n = 6, √ n−1 . which are isomorphic (as groups) to S5 , A5 , S6 or A6 , and k(S) ≤ 3 Consequently first claim that if |S : H| = 2 or |H b > 2 and d > 2.

S is not primitive. Let T be a point stabilizer in S. We H is a proper intermediate group, T ⊂ H ⊂ S, then : T | = 2. Let |S : H| = b and |H : T | = d, and assume Let C = CoreS (H) and |C : C ∩ H| = |CT : H| = d0 ,


177

|S : CT | = b0 . Then C has b0 orbits on Ω = {1, · · ·, n}, each of size d0 . Of course S/C is a transitive permutation group of degree b, 3 ≤ b < n, and √ b−1 . If d0 > 2, then by assumption and (1.7b) so k(S/C) ≤ 3 k(S) ≤

√ (d0 −1)b0 √ b−1 √ n+b−b0 −1 √ n−1 3 · 3 ≤ 3 ≤ 3 ,

because n = b0 d0 and b0 ≥ b. Thus d0 = 2 and b0 = n2 . It follows that n C is an elementary abelian 2-group, with k(C) = |C| = 2 2 , and CT ⊂ H 1 √ √ 1 (d > d0 = 2). Hence b ≤ n4 . Since 2 2 · 3 4 < 3, we get the desired result. So letting H be a proper intermediate group, the following three possibilities have to be considered: (i) T is maximal in H and |S : H| = 2; (ii) H is maximal in S and |H : T | = 2; (iii) There is a further intermediate subgroup X, such that T ⊂ H ⊂ X ⊂ S say, and then H is maximal in X and X is maximal in S, and |H : T | = 2 = |S : X|. Case (i): Let d = |H : T |, and let N = CoreH (T ). Suppose first that d ≥ 7. Observe that H/N is a primitive permutation group of degree √ d d. Hence k(H/N ) ≤ 2 by Lemma 9.4a. For any x ∈ S r H we have √ m−1 by induction N ∩ N x = 1 and k(N ) = k(N x ) = k(N · N x /N ) ≤ 3 √ d+2 √ d−1 · 3 . Since we are (d ≥ 3). Thus by k(H) ≤ 2 · k(H/N ) · k(N ) ≤ 2 √ d √ 2d−1 √ d+2 , we have 2 > 3 . But this is false assuming that k(S) > 3 for d ≥ 4 (even). Thus d ≤ 6. Note that p = 7 when S is not solvable. Let d = 6. Then by inspection √ 11 k(N ) ≤ k(H/N ) ≤ 11 and k(S) ≤ 2 · k(H/N ) · k(N ) ≤ 242 < 3 . If √ 9 d = 5 then k(H/N ) ≤ 7, k(N ) ≤ 7 and k(S) ≤ 98 < 3 . For d ≤ 4, H and hence S are solvable. Consider d = 3. Then |S| ≤ 2|H/N |2 ≤ 72. √ 2d−1 ≥ 16 by assumption, we have Since k(S3 wr S2 ) = 9 and k(S) > 3 |S| ≤ 36, even |S| = 36 (for otherwise S were abelian). Then |H| = 18 and |N | = 3, and k(S) ≤ 3 · 3 · 2 = 12, a contradiction. Let d = 4. Then by inspection k(H/N ) ≤ 5, and k(N/N ∩ N x ) ≤ 5 for x ∈ S r H. This gives k(S) ≤ 50, However, we are assuming that √ 7 k(S) > 3 , hence k(S) ≥ 47. This forces that k(H/N ) = k(H/N x ) = 5. Since H/N and H/N x are isomorphic to primitive permutation groups of

178


degree 4, we see that H/N ∼ =N ∼ = S4 . We conclude that S ∼ = S4 wr S2 and k(S) = 20. Case (ii): Let b = |S : H|, and let N = CoreS (H). Here S/N is a primitive permutation group of degree b, and N is an abelian √ elementary m 2-group of order at most 2b . If b ≥ 7 then k(S/N ) ≤ 2 by Lemma 9.4a. √ +2b √ 2b−1 . Since by assumption k(S) > 3 this It follows that k(S) ≤ 2 √ 3b √ 2b−1 and b < 10. forces that 2 > 3 Let b = 9. If S/N is solvable, then S/N is of affine type and so k(S/N ) ≤ 11 as seen in the proof of Lemma 9.4a. This gives the result. If S/N is a nonsolvable (primitive) 50 -subgroup of S9 , then it is isomorphic to a (subnormal) subgroup of L2 (8) : 3 and so even k(S/N ) ≤ 9 [Atlas]. There is no nonsolvable primitive 70 -subgroup of S9 . Let b = 8 or 7. If S/N is solvable (of affine type), then k(S/N ) ≤ b, which is as required. If S/N is a nonsolvable, primitive p0 -subgroup of S8 , then p = 5 and S/N is isomorphic to L3 (2) or to L3 (2) : 2 and k(S/N ) ≤ 9, or it is the affine group 23 : L3 (2) of degree 8. We know already that k(23 : L3 (2)) = 11. We get the desired contradiction here, as well as for b = 7 where k(S/N ) = k(L2 (7)) = 6. Let b = 6. Then S is solvable or p = 7 and k(S/N ) ≤ 11, yielding the result. Let b = 5. If S/N is a nonsolvable (70 -) subgroup of S5 , then it is S5 or A5 and so k(S/N ) = 7 or 5. From the [Atlas] we infer that S is isomorphic to a subgroup of 25 : S5 = Z2 wr S5 , which has just 36 conjugacy classes. Since 25 is the natural permutation module for S5 over F2 , this may be computed by means of formula (1.10b) and shows that S contains 24 : A5 . √ 9 At any rate, k(S) ≤ 36 < 3 . The remainder is straightforward. Case (iii): Let |X : H| = m and N = CoreX (H), M = CoreX (T ). Then M ∩M x = 1 for any x ∈ SrX, N/M is an elementary abelian 2-group of order at most 2m , and X/N is a primitive permutation group of√degree m m (and order prime to p). Suppose that m ≥ 7. Then k(X/N ) ≤ 2 by √ 3m √ 2m−1 (9.4a), hence k(X/M ) ≤ 2 . Further k(M ) = k(M x M/M ) ≤ 3 √ 3m √ 2m−1 · 3 . Since k(S) > by induction. It follows that k(S) ≤ 2 · 2 √ 4m−1 3 , this forces that m < 12. If X/N is solvable (of affine type), one even obtains m ≤ 9, and these cases are easily treated. So let X/N be not solvable in what follows. Note that k(X/N ) > 3m /2m+1 by assumption.


179

Let m = 11. Then p = 7 and either X/N ∼ = L2 (11) and k(X/N ) = 8, or X/N ∼ M and k(X/N ) = 10. We obtain the desired contradiction. = 11 Let m = 10. Then p = 7 and X/N either is a (primitive) subgroup of A6 .2 , in which case k(X/N ) ≤ 22, or of S5 (in its primitive action on the 2-subsets of {1, · · ·, 5}). 2

The possibilities m = 9, 8, 7 are ruled out using the information given in Case (ii). For m = 6 we have X/N ∼ = S6 or A6 , but we have to argue more carefully. Choose x ∈ S r X. Then N · M x is a subnormal subgroup of X. Thus either M N x ⊆ N and k(M ) = k(M M x /M ) ≤ 26 , or N M x /N is S6 or A6 , that is, N M x is a subgroup of X of index 1 or 2. The latter case cannot happen since 1 = M ∩ M x = M ∩ (M x ∩ N ) and so M ∼ = M (M x ∩ N )/(M x ∩ N ) would be an elementary abelian 2-group (but M x √ 24−1 . The would not). Using (1.10b) we get k(S) ≤ 2 · 11 · 26 · 26 < 3 ¤ same argument works for m = 5. This completes the proof. Examples 9.4c. The upper bound given in Lemma 9.4a can be improved considerably for large degree n. For instance, in [Gluck et al., 2004] it has been shown that if S is a primitive permutation 50 -group of degree n > 20, n then k(S) ≤ 2 5 . On the other hand, there are examples (of imprimitive groups) where one has proper lower bounds. (i) Let S = S4 wr Zm for some integer m ≥ 2, which is a transitive (but imprimitive) permutation group of degree n = 4m. Since the base group of the wreath product is a direct product of m copies of S4 and has index m m n in S, by part (i) of Theorem 1.7a k(S) ≥ 5m = n4 5 4 . By Proposition 8.5d, n 5n k(S) ≤ 5 4 + 4 − 5. (ii) Let S = Zp wr Zp for some prime p, which is a (transitive) Sylow p-subgroup of Sn , n = p2 . By Proposition 8.5d, k(S) = pp−1 + p2 − 1 > √ log2 p ( n−1) . 2 (iii) Let S be a Sylow p-subgroup of Sn where n = pm+1 for some prime p and some m ≥ 2. One knows that S ∼ = Zp wr Zp wr · · · wr Zp with 1 m−1 m + 1 occurrences of Zp . By induction one gets that k(S) ≥ p p−1 · bp n where logp b = (p − 2)/(p − 1). So for instance k(S) > 3 6 when p = 3.

Chapter 10

The Final Stages of the Proof In proving the k(GV ) theorem we may assume that G is irreducible on V (Proposition 3.1a). By the Robinson–Thompson Theorem 5.2b we may assume that there is no real vector in V for G. Then, by the Riese–Schmid Theorem 8.4, the pair (G, V ) is nonreal induced. The proof is accomplished now using that the class numbers for the nonreal reduced pairs are rather small, and using the Liebeck–Pyber Theorem 9.3 and its improvement. This actually works except for one case (p = 5), where some additional considerations are needed. 10.1. Class Numbers for Nonreal Reduced Pairs Throughout we let (H, W ) be a nonreal reduced pair inducing the pair (G, V ). Let E be the core of H. Proposition 10.1. Let Y be a subnormal subgroup of H. (i) If Y ⊇ E then k(Y W ) < |W |/2 unless (H, W ) is of type (Q8 ) for p = 5 or p = 7. For the type (Q8 ) and p = 7 (where |W | = 49) we have k(Y W ) ≤ k(HW ) = 27. (ii) Let p = 3 (where (H, W ) is of type (25− ) and |W | = 81) and let Y ⊇ Z(E). Then k(Y W ) ≤ k(Z(E)W ) = 42. (iii) Let p = 5 (where (H, W ) is of type (Q8 ) and |W | = 25) and let Y 6= 1. Then k(Y W ) ≤ k(HW ) = 20. Proof. We make use of Proposition 3.1b and knowledge of the orbit structure of H on W , including knowledge of the point stabilizers (Secs. 6.1 and 7.1). The class numbers k(H), k(Y ) are computed using the Clifford– Gallagher formula (1.10b). We carry out only a few cases. The assumption that Y is subnormal in H is crucial. Let Z = Z(E) and U = E/Z. Type (Q8 ) : Let p = 5. Then H ∼ = (Q8 ◦ Z4 ).S3 . There are k(S3 ) = 3 irreducible characters of H lying over each of the two faithful irreducible characters of Y4 = Q8 ◦ Z4 (extendible to H since M(S3 ) = 1), and over 180

The Final Stages of the Proof

181

each of its two S3 -invariant linear characters. (We use the notation which will be introduced in Lemma 10.4a below.) Two further linear characters of Y4 are stable under a fixed involution in S3 , each having two extensions to this inertia group and induce up to H. Thus k(H) = 3 · 2 + 3 · 2 + 2 · 2 = 16. Of course k(Q8 ) = 5, k(Y4 ) = 10 and k(H 0 ) = 7 (H 0 = Y3 ∼ = SL2 (3)). For the subgroup Y = Y2 of H with index 2 we have k(Y2 ) = 14. Now use that H is transitive on W ] and that CH (w) ∼ = Z4 for any w ∈ W ] . We obtain that k(HW ) = 20 and k(Y2 W ) = 16. The cases p = 7, 13 are treated similarly. ¯ ¯ Type (25− ) : Let p = 3. Then H ∼ = (21+4 − ) : H where H = Z5 : Z4 . There ¯ are k(H) = 5 irreducible characters of H over the unique faithful irreducible ¯ = 1). Since Z5 is irreducible on U , Z4 character of E ∼ (as M(H) = 21+4 − ¯ : Z2 | = 10 nonsingular fixes a unique singular point in U , and since the |H ¯ ¯ = 16. points have stabilizer Z2 in H, we get k(H) = 5 · 1 + 4 · 1 + 2 · 1 + k(H) ] ∼ Using that H is transitive on W and that CH (w) = Z8 for each w ∈ W ] we get k(HW ) = 16 + 8 = 24. Of course k(E) = 17, and E has 5 orbits on W ] of size 16, whence k(EW ) = 17 + 5 · 2 = 27. Also Z = Z(E) has 40 orbits on W ] of size 2, so k(ZW ) = 42. Observe that if Y is generated by a noncentral involution in E, then k(Y W ) = 54. So the assumption Y ⊇ Z(E) cannot be omitted. ¡ ¢ Type (33+ ) : Let p = 7 and H ∼ : Sp2 (3) × Z2 . There are three = 31+2 + (2) H-orbits on W ] with point stabilizers Z6 , Z2 × Z3 and Sp2 (3). Hence k(H) = 2(2 · 7 + 1) = 30 and k(HW ) = 30 + 6 + 18 + 7 = 61 (whereas |W | = 73 ). Similar computation for the subnormal subgroups Y containing E∼ = 31+2 + . Type (2.A5 ) : Let p = 31. Then H ∼ = 2.A5 × Z15 and k(H) = 9 · 15 = 135. There are two orbits of H on W ] with point stabilizers Z3 , Z5 , hence k(HW ) = 135 + 3 + 5 = 143 (whereas |W | = 312 ). The group E = 2.A5 has 8 regular orbits on W ] , hence even k(EW ) = 9 + 8 = 17. Type (2.A6 ) : Here H ∼ = 2± S6 × Z3 and p = 7, and we have three H-orbits ] on W with stabilizers Z3 , Z6 and Z3 wr S2 . Apply Proposition 3.1b. ∼ Sp (3) × Z3 and k(H) = 3 · 34 = 102 Type (Sp4 (3)) : Let p = 7. Then H = 4 (Atlas). There are two orbits on W ] with point stabilizers SL2 (3) × Z3 and (31+2 : Q8 ) × Z3 , which have class numbers 21 and 33, respectively. Hence + k(HW ) = 156 (whereas |W | = 74 ). The remainder is done similarly. ¤

182


10.2.

Counting Invariant Conjugacy Classes

By assumption we have an imprimitivity decomposition V = W1 ⊕ · · · ⊕ Wn for some integer n ≥ 1, where the pairs (NG (Wi )/CG (Wi ), Wi ) are isomorphic to (H, W ) and permuted transitively by G. From the classification of the nonreal reduced pairs in Chapters 6 and 7 it follows that V is an absolutely irreducible Fp G-module, and that p ∈ {3, 5, 7, 11, 13, 19, 31}. Let us write Gi = NG (Wi ) and Hi = Gi /CG (Wi ) for each i, and let Tn N = i=1 Gi be the normal core. Let E ∼ = Ei be the cores of (H, W ) ∼ = (Hi , Wi ). Identify (H, W ) with (H1 , W1 ) say (as reduced pairs). Recall that distinct normal subgroups of H are not isomorphic (and hence are characteristic in H). Let S = G/N act (transitively) on the set Ω = {1, · · ·, n} like on the {Wi }. Because of Proposition 10.1 we may assume that n ≥ 2. The degree n of the permutation group S is also called the degree of (G, V ). Since G is faithful on V , we may identify G with a subgroup of the wreath product Hwr S (Lemma 8.2a). Let Ti = CN (Wi ) and Ni = N/Ti for i ∈ Ω, and denote by N0 ∼ = Ni their common image in L T H = H0 . Let Ri = j6=i Tj = CN ( j6=i Wj ) so that R = R1 × · · · × Rn is a subgroup of N whose direct factors are permuted transitively by G (or S). Let R0 ∼ = Ri ∼ = RCG (Wi )/CG (Wi ) denote the common image in H. We know from Theorem 8.5c that R0 ⊇ E except possibly when p = 3 or p = 5. In the exceptional cases at least R0 ⊇ Z(E), and always N0 ⊇ E. Of course R is normal in G, and both N0 and R0 are characteristic normal subgroups of H = H0 (R0 ⊆ N0 ). We let Q = N/R and Q0 = N0 /R0 . We need some bounds for the number kg (N ) = |CC`(N ) (g)| of conjugacy classes of N fixed by some element g ∈ G. Let us call a vector v ∈ V an n-vector provided its support is Ω, that is, v = w1 + · · · · +wn with all components wi ∈ Wi being nonzero. Proposition 10.2a. Suppose S = hN gi is cyclic (g ∈ G). Then: (i) kg (N ) ≤ min{|N0 |, k(R0 ) · kg (Q)}. (ii) kg (N V ) ≤ min{|N0 W |, k(R0 W ) · kg (Q)}. P (iii) kg (N V ) ≤ kg (N ) + j k(CN (vj )) where {vj } is a set of representatives of the N -orbits on V ] consisting of n-vectors.


183

Proof. For later use we describe exactly the assumptions needed for the argument. Of course g normalizes N and R, acts on Q and permutes the {Ri } (and {Hi }) cyclically. The nth power g n normalizes each Ri (since g n ∈ N or, what is essential, since g n induces the identity permutation on the set {Ri }). There is a corresponding embedding of GV into the wreath product (HW )wr S, having base group BV = H1 W1 × · · · × Hn Wn . Here RV = R1 W1 × · · · × Rn Wn is a normal subgroup of N V , with quotient group Q = (N V )/(RV ), and g leaves N V and RV invariant and permutes the {Ri Wi } cyclically. Therefore (ii) is immediate from (i). Let us prove (i). For each y ∈ N , CN (gy) maps onto a subgroup of CQ (gy) with kernel CR (gy). Hence |CN (gy)| ≤ |CQ (gy)|·|CR (y)|. By (1.7c) therefore 1 X 1 X |CN (gy)| ≤ |CQ (gy)| · |CR (gy)|. kg (N ) = |N | |N | y∈N

y∈N

Now CN (gy) ∩ CN (W1 ) = 1 for each y ∈ N , because gy permutes the {Wi } cyclically and y leaves each Wi invariant. Hence CN (gy) maps injectively into N1 = N/CN (W1 ) ∼ = N0 and so |CN (gy)| ≤ |N0 |, for all y ∈ N . Consequently kg (N ) ≤ |N0 |. P Let α = |N1 | y∈N |CQ (gy)|. Let {tj } be a transversal to R in N , and |R| 1 let g¯ = Rg. Since CQ (gtj x) = CQ (gtj ) for each x ∈ R, and since |N | = |Q| , we get 1 X |CQ (¯ α= g y¯) = kg (Q), |Q| y¯∈Q P 1 again by Eq. (1.7c). Let β be the maximum of the |R| x∈R |CR (gtj x)|, with tj varying over the transversal, the maximum being obtained for tj = t0 say. The estimate in the preceding paragraph shows that kg (N ) ≤ α · β. So we have to show that β ≤ k(R0 ). Let x = x1 · · · xn and h = h1 · · · hn be elements of R (with xi , hi ∈ Ri ). Assume g, and hence gt0 and g0 = gt0 (x/x1 ), map onto the cycle (12...n) in S. Then h is centralized by (g x )n 0 x1 gt0 x = g0 x1 if and only if hg10 x1 = h2 , · · ·, hgn−1 = hn and h1 = h1 0 1 . g n−1 g n−2

(g x )n

gn x

Now (g0 x1 )n = g0n x10 x10 · · · x1 and so h1 0 1 = h10 1 , because h1 is centralized by Ri for i 6= 1. We conclude that |CR (gt0 x)| = |CR (g0 x1 )| ≤ |CR1 (g0n x1 )|. Just using that g0n normalizes R1 , as before 1 X |CR1 (g0n x1 )| = k(g0n ) (R1 ) ≤ k(R1 ) = k(R0 ). |R1 | x1 ∈R1

184


P |R| = |R0 |n−1 eleThus |R11 | x1 ∈R1 |CR (gt0 x1 x2 · · · xn )| ≤ k(R0 ) for all |R 1| P 1 ments (x2 , ···, xn ) ∈ R2 ×···×Rn . Consequently β = |R| x∈R |CR (gt0 x)| ≤ k(R0 ). It remains to prove (iii). C`(N V ) and Irr(N V ) are isomorphic hgisets by Brauer’s permutation lemma (Theorem 1.4b). The partition V = U U j vj N into N -orbits corresponds to a partition Irr(N V ) = j Irr(N V |λj ), with CN (vj ) = IN (λj ) when the λj ∈ Irr(V ) are chosen properly (Proposition 3.1b). These partitions are preserved by g since Irr(N V |λj )g = Irr(N V |λgj ) for each j. Thus kg (N V ) is the sum of the fixed points under g in Irr(N V |λj ), the sum taken over those λj corresponding to hgi-invariant N -orbits on V . One such summand is kg (N ), the number of fixed points under hgi of Irr(N V |1V ) = Irr(N ). Since g permutes the {Wi } cyclically, only those N -orbits on V ] can be hgi-invariant which are represented by n-vectors. Hence the result. ¤ We would like to have a substitute for Proposition 10.2a in the general Q case. Let g ∈ G. If B = J BJ is a decomposition of the base group according to the cycle decomposition of the permutation g¯ = N g in S, Q the natural projection map N → J BJ is a hgi-homomorphism, and it is injective. However, letting NJ be the image of N in BJ , the induced map Q C`(N ) → J C`(NJ ) need not be injective. (An enlightening example is S3 ∆Z2 S3 .) So the map on the fixed point sets under hgi need not be injective likewise. The following lemma is used only in the very last step (p = 5) of the proof of the k(GV ) conjecture (and for N V in place of N ). Lemma 10.2b. Let g ∈ G. According to the the cycle decomposition of Ω under the permutation g¯ = N g in S, with orbits J, let NJ be the image Q Q of N in BJ = i∈J Hi . Then RJ = i∈J Ri is a normal subgroup of NJ , and both RJ and NJ are stable under g. For each J let kJ be the maximum of the |CC`(Y ) (gt)| where Y varies over all subgroups of NJ containing RJ Q which are normalized by gt for some t ∈ N . Then kg (N ) ≤ J kJ . Proof. Let J1 , ···, Jr be the distinct h¯ g i-orbits on Ω. We argue by induction on r. For r = 1 the assertion is trivial. So let r > 1. Let M be the image of Qr−1 N under the hgi-projection of B onto i=1 BJi with respect to BJr . Since the maps N ³ NJi , 1 ≤ i ≤ r − 1, factor through N ³ M , the inductive Qr−1 hypothesis applies showing that kg (M ) ≤ i=1 kJi . M M N Fix ym ∈ M such that ym ∈ CC`(M ) (g). Of course ym = ym . Let Y be the image of CN (ym ) under the hgi-projection of B onto BJr . Then


185

gt Y ⊇ RJr . There exists t ∈ N such that ym = ym . Since gt centralizes ym and normalizes N , it normalizes CN (ym ) and Y (as the projection is compatible with the action of gt). Let y N be a hgi-invariant conjugacy class M of N which maps onto ym . We may replace y ∈ N by an N -conjugate, if necessary, such that y maps onto ym . Then y ∈ CN (ym ), and we may write uniquely y = ym yr = yr ym where yr is the image of y in Y . There exists t0 ∈ N such that y gt = y t0 , and we have t0 t0 ym yrgt = y gt = y t0 = ym yr .

We conclude that ym = y t0 and yrgt = yrt0 (by uniqueness), hence t0 ∈ CN (ym ) and yrY ∈ CCl(Y ) (gt). Now let yeN be another hgi-invariant conjugacy class of N mapping M onto ym , and choose ye such that ym is its component in M . Write uniquely ye = ym yer = yer ym with yer ∈ Y and yerY ∈ CC`(Y ) (gt), as before. If yerY = yrY , then yeCN (ym ) = ym yerY = ym yrY = y CN (ym ) and so yeN = y N . The result follows since, by definition, kgt (Y ) ≤ kJr . 10.3.

¤

Nonreal Induced Pairs

Theorem 10.3 (Riese–Schmid). Let (G, V ) be properly induced (n ≥ 2) from the nonreal reduced pair (H, W ). Then k(GV ) ≤ 21 |V |, except possibly when p = 5. Proof. We use the notation introduced in Sec. 10.2. In addition let Yn = Tn = N/CN (Wn ) and let Yi = CN (Wi+1 ⊕ · · · ⊕ Wn )/CN (Wi ⊕ Wi+1 ⊕ · · · ⊕ Wn ) for i = 1, · · ·, n − 1. Since CN (Wi+1 ⊕ · · · ⊕ Wn ) ⊇ Ri and Ri ∩ CN (Wi ⊕ · · · ⊕ Wn ) = 1, Yi may be identified with a subnormal subgroup of Hi containing Ri . So Wi is a faithful Fp Yi -module. Suppose first that p 6= 3 (and p 6= 5). Then Yi ⊇ Ri ⊇ Ei by Theorem 8.5c. Hence k(Yi Wi ) < 21 |W | by Proposition 10.1 (i) unless (H, W ) is of type 27 |W |. (Q8 ) with p = 7. In this exceptional case we have k(Yi Wi ) ≤ 27 = 49 Repeated application of (1.7b) yields that k(N V ) ≤

n Y i=1

k(Yi Wi )
3. This is less than 0.502 for n = 2, 3. Note that ( 49 27 2 n = 3, and ( 49 ) · 2 ≤ 0.61. ∼ 2− S4 ×Z3 We have to improve the last bounds (for n = 2, 3). Here H = (p = 7). By Theorem 8.5c, R0 ⊇ E ∼ = Q8 . Let n = 2, and let S = hN gi be of order 2. Then Q = N/R is a quotient group of S3 × Z3 . Hence k(Q) ≤ 3 · 3 = 9, and if k(Q) = 9 then R0 ∼ = Q8 and k(R0 W ) = 11. It is easy to check that k(Q) · k(R0 W ) ≤ 99 in each case. We know that 27 2 ) |V | = 729. By Proposition 10.2a (ii) at most kg (N V ) ≤ 99 k(N V ) ≤ ( 49 of the at most 729 conjugacy classes of N V are hgi-invariant. By Theorem 1.4b at most 99 of the at most 729 irreducible characters of N V are stable in GV . By (1.10b) therefore k(GV ) ≤ 2 · 99 + (729 − 99)/2 = 513, which is less than 41 |V |. For n = 3 we have to examine the cases where S ∼ = A3 or S3 . Use that k(N V ) ≤ 27 3 and that |H| · |W | = 3 · 48 · 27 conjugacy classes of N V are fixed by a 3-cycle in S by Proposition 10.2a. Let p = 3. Then |W | = 34 and k(Yi Wi ) ≤ 42 by Proposition 10.1. By √ n−1 for n ≥ 3. Hence Proposition 9.4b, k(S) ≤ 3 k(GV ) ≤ k(N V ) · k(S) ≤

¡ 42 ¢n √ n−1 1 3 |V | ≤ |V |, 81 2

42 2 ) ·2 < except when n = 2, in which case k(GV ) ≤ k(N V ) · 2 ≤ ( 81 0.6|V |. Again this bound is easily improved. The worst case happens when R0 ∼ = Z(H) has order 2, in which case Q ∼ = H/Z(H). Then k(R0 W ) = 42 and k(Q) = 12. Hence by Proposition 10.2a at most 42 · 12 = 504 of the k(N V ) ≤ 422 irreducible characters of N V are invariant in GV . Application of (1.10b) gives k(GV ) ≤ 2 · 504 + (422 − 504)/2 = 1.638, which is less that 41 |V | (|V | = 812 ). The proof is complete. ¤

10.4.

Characteristic 5

It remains to examine the case where (H, W ) is the nonreal reduced pair of type (Q8 ) with p = 5. Having established Theorem 10.3 in 2001, published


187

in [Riese–Schmid, 2003], it took almost two years to handle this case too. The thorough analysis made in [Gluck et al., 2004] yields in this case the estimate k(GV ) ≤ 21 |V | as before (n ≥ 2). In what follows we argue, using some ideas from [Keller, 2006], on the basis of Proposition 10.2a and Lemma 10.2b. This makes the approach considerably shorter (though the result is weaker). We use the notation introduced in Sec. 10.2. For convenience of the reader we give some additional information on the group H, which is a 5-complement in GL2 (5) (uniquely determined up to conjugacy). As already seen in Sec. 3.2, H is transitive on the (2) nonzero vectors of the standard module W = F5 . Also, H is the standard holomorph of Q8 ◦ Z4 . Hence H/Z(H) ∼ = S4 , but the two covering groups 2± S4 are not involved in H (nor a Singer cycle). Lemma 10.4a. Let (H, W ) be of type (Q8 ) with p = 5. There are up to conjugacy just 11 subnormal subgroups Ya of H = Y0 containing the centre of the core E of H (listed below). The numbering is such that whenever Ya contains some conjugate of Yb then a ≤ b. Ya

k(Ya )

Y0 = H 16 Y1 = O2,3 (H) 14 Y2 = H 0 ∼ 7 = SL2 (3) Y3 = O2 (H) ∼ = Q8 ◦ Z4 10 Y4 = E = H 00 5 Y5 ∼ 5 = D8 Y6 ∼ 8 = Z4 × Z2 Y7 = Z(H) ∼ 4 = Z4 Y8 ∼ 4 = Z4 Y9 ∼ 4 = Z2 × Z2 Y10 = Z(E) 2

k(Ya W ) 20 16 8 16 8 14 14 10 10 16 14

k(Ya /Y11 ) 10 8 4 8 4 4 4 2 2 2 1

The statements are easily verified (arguing as for Proposition 10.1). The normal subgroups of H are determined by their order. The subnormal but not normal subgroups of H are of type Y5 , Y6 , Y8 and Y9 , each having 3 conjugates under H. For each subgroup Y of H we have k(Y W ) ≤ 20, except for two conjugacy classes of abelian groups Y = CH (w) respectively CH (w) × Z(H) for some w ∈ W ] , where k(Y W ) = |W |. If Y 6= H is a nonabelian subgroup, then k(Y W ) ≤ 16 except when Y ∼ = Z4 wr Z2 is a Sylow 2-subgroup of H (in which case k(Y W ) = 20).

188


It is also easy to compute k(Ya /Yb ) whenever Yb is a normal subgroup of Ya . Knowing that |CYa (w)| = 1, 2 or 4 for each w ∈ W ] , which may be deduced from the first row of the table, application of Proposition 3.1b enables us to compute from k(Ya W ) − k(Ya ) the number of Ya -orbits on W ] . The subgroups of H of order greater than or equal to 8 have at most 5 orbits on W ] . (By direct computation or using [GAP] one can establish the complete orbit structure for each subgroup of H.) Lemma 10.4b. Let (H, W ) be of type (Q8 ) with p = 5, and let n = 2. Then k(GV ) ≤ 21 |V |. Proof. In this case S = hN gi has order 2. Also G1 = N = G2 so that N = H1 ∆Q H2 is a fibre-product of the Hi ∼ = H amalgamating Q = N/R (R = R1 ×R2 ). We also have an extension R0 ½ H ³ Q. As in the proof of Theorem 10.3 we have k(N V ) ≤ k(N1 W1 ) · k(R2 W2 ) ≤ 202 (R2 = CN (W1 ) and N1 = N/R2 ∼ = H). Suppose G = Hwr S, that is, R0 = H = Y0 . Then GV = (HW )wr S and so by Proposition 8.5e and Lemma 10.4a k(GV ) = 2 · k(HW ) + k(HW )(k(HW ) − 1)/2 = 230. Suppose R0 = Y1 , Y2 , Y3 , Y4 or Y7 . Application of Proposition 10.2a and of Lemma 10.4a yields that kg (N V ) ≤ k(R0 W ) · k(Q) = 32, 32, 48, 48, 50, respectively. In the worst case, in view of Theorem 1.4b, at most 50 of the at most 202 irreducible characters of N V are stable in GV . On the basis of the Clifford–Gallagher formula (1.10b) we therefore obtain the estimate k(GV ) ≤ 2 · 50 + (400 − 50)/2 = 275. It remains to examine the case where R0 = Y10 . Then kg (N ) ≤ 2 ·10 = 20 by part (i) of Proposition 10.2a, and part (ii) gives kg (N V ) ≤ k(R0 W ) · k(Q) = 14 · 10. We improve this latter bound by applying part (iii) of this proposition. We know that N1 is transitive on W1] and that CN (w1 )/R2 is cyclic of order 4 for each w1 ∈ W1] . Since R1 = CN (W2 ) is fixed point free on W1] , CN (w1 ) ∩ R1 = 1 and so CN (w1 ) is faithful on W2 (of order 8). Similarly, CN (v) = CN (w1 ) ∩ CN (w2 ) intersects R2 trivially for each w2 ∈ W2] , whence |CN (v)| = 1, 2 or 4. We also know from Lemma 10.4a that CN (w1 ) has at most 5 orbits on W2] . (Either CN (w1 ) is cyclic with


189

3 regular orbits on W2] , or it corresponds to Y9 having 4 orbits, or it is of type CH (w) × Y10 for some w ∈ W2] , in which case one has 5 orbits.) Thus kg (N V ) ≤ kg (N ) + 5 · 4 ≤ 40. Noting that k(N V ) ≤ 20 · 14 = 280 here, formula (1.10b) yields that k(GV ) ≤ 2 · 40 + (280 − 40)/2 = 200. ¤ The last estimate can be easily improved. In fact, in Lemma 10.4b the upper bound k(GV ) ≤ 152 holds unless G = Hwr S2 . Proposition 10.4c. Let (H, W ) be of type (Q8 ) with p = 5, inducing the Sn nonreal pair (G, V ) as before (n ≥ 2). Then for any g ∈ G r i=1 Gi we 2 have kg (N V ) < |V | 3 . Proof. By hypothesis g¯ = N g is a permutation in S = G/N without fixed points on Ω. According to the cycle decomposition of g¯ we obtain a decomposition of the base group BV of (HW )wr S. By virtue of Lemma 10.2b we may replace g¯ by one of its cycles, in the following sense: Suppose g ∈ (HW )wr S permutes W1 , · · ·, Wn cyclically and leaves N V and RV invariant. Let Y be any subgroup of N V containing RV which is 2 normalized by g. Then prove that |CC`(Y ) (g)| ≤ |V | 3 . The crucial point is that Y ⊇ RV . In particular V = O5 (Y ) and Y /V is a 50 -group. Note that RV is normal in Y as Y ⊆ N V . By the Schur– e in Y and, replacing g by Zassenhaus theorem we find a 5-complement N e . Then gv for some v ∈ V , if necessary, we may assume that g normalizes N e e e R = N ∩ RV is a g-invariant normal subgroup of N which is isomorphic to ei = C (L Wj ) is isomorphic to Ri for each i. R. In fact R j6=i e N e = N, R e = R and R ei = Ri . Then, as By abuse of notation we write N before, R is a g-invariant normal subgroup of N mapping onto the normal subgroup R0 ∼ = Ri of H = Y0 , with R0 ⊇ Y10 in the notation of Lemma 10.4a. Let Ti = CN (Wi ) and Ni = N/Ti for each i, as before, which are permuted cyclically by g. In contrast to our previous experience the Ni can be, at first, map onto subgroups of H which are not normal, possibly not even subnormal. But they all contain R0 . So we may define again Q0 = N0 /R0 (where N0 = N1 according to our convention). In addition Q = N/R is a g-invariant quotient group as before. We do not know whether g n is an element of the (new) group N (or N V ), but g n

190


normalizes each Ni and each Ri . Consequently Proposition 10.2a applies. In particular kg (N V ) ≤ |N0 W | = |N0 | · |W | ≤ 96 · 25 = 2.400. 1

n

2

This is less than |V | 2 = |W | 2 if n ≥ 5. For n = 4 we have b|V | 3 c = 5.343, 2 and |V | 3 = 625 for n = 3. Therefore we only have to consider the cases n = 2, 3. 2

n = 2: Then b|V | 3 c = 73. Keep in mind that N0 may be any subgroup of H = Y0 containing R0 ⊇ Y10 (unlike the situation in Lemma 10.4b). Noting that T1 = R2 , T2 = R1 and R = R1 × R2 we see that Q0 = Q and that N = N1 ∆Q N2 is a fibre-product. We know from Lemma 10.4a that k(R0 W ) ≤ 20, and that kg (N V ) ≤ k(Q) · k(R0 W ) by Proposition 10.2a. If k(R0 W ) = 20, then R0 = H, Q = 1 and kg (N V ) = 20. We may thus assume that k(R0 W ) ≤ 16, and that k(Q) > 4. Inspection of Lemma 10.4a yields that R0 must be contained in Y3 ∼ = Q8 ◦ Z4 . We distinguish two cases. Case 1: Q is a 30 -group Then N0 is a 2-group, hence contained in a Syow 2-subgroup P ∼ = Z4 wr S2 of H. Note that P 0 = Y8 (up to conjugacy in H). Since R0 ⊇ Y10 is normal in H, we have k(Q) · k(R0 W ) < 73 unless R0 = Y10 and either N0 = P or N0 = Y3 (Lemma 10.4a). Let us consider the (worse) case where N0 = P . Then k(Q) = 10, |R0 | = 2 and k(R0 W ) = 14. Hence kg (N ) ≤ k(Q) · k(R0 ) = 20 by part (i) of Proposition 10.2a, and kg (N V ) ≤ 140 by part (ii). This will be improved by using part (iii) of this proposition. One checks that N0 = P has two orbits on W ] , with point stabilizers of order 2 and 4. Let v = w1 + w2 be a 2-vector in V (wi ∈ Wi] for i = 1, 2). Then CN1 (w1 ) = CN (w1 )/R2 has order 2 or 4. Since R1 = CN (W2 ) is fixed point free on W1] , CN (w1 ) is faithful on W2 (of order 4 or 8). Similarly, CN (v) ∩ R2 = 1 and so |CN (v)| is a divisor of 2 or of 4. If CN (w1 ) is cyclic, it has 6 regular orbits on W2] when |CN (w1 )| = 4, and 3 regular orbits otherwise. If CN (w1 ) is not cyclic and of order 4, it is of type Y9 and has 4 regular orbits and 4 orbits of size 2. If |CN (w1 )| = 8 we have at most 8 orbits (Lemma 10.4a). We conclude that there are at most 5 orbits of N on 2-vectors of V with point stabilizers CN (v) of order dividing 4, and at most 8 such orbits with stabilizers of order dividing 2. Hence kg (N V ) ≤ kg (N ) + 5 · 4 + 8 · 2 ≤ 56,


191

as desired. The case where N0 = Y3 is treated similarly: The number of N -orbits on V consisting of 2-vectors is at most 3 · 8 = 24, with point stabilizers of order 2 or 1. This yields that kg (N V ) ≤ kg (N ) + 24 · 2 ≤ 64 in this case. Case 2 : |Q| is divisible by 3 Then |N0 | ≥ 24. There are two conjugacy classes of subgroups in H which act regularly on W ] , namely Y3 ∼ = SL2 (3), which is normal in H, and the 3-Sylow normalizers Y ∼ S × = 3 Z4 in H. Since k(Q) > 4 and R0 ⊇ Y10 , we only have to treat the second case, with N0 = Y and R0 = Y10 . Then k(Q) = 6 and k(R0 W ) = 14, and kg (N ) ≤ 12, kg (N V ) ≤ 6 · 14 by Proposition 10.2a. We have to improve the latter bound. For any 2-vector v = w1 + w2 the stabilizer CN (w1 ) = R2 has 12 regular orbits on W2] , so that we get kg (N V ) ≤ kg (N ) + 12 · 1 ≤ 24. It remains to examine the cases where N0 = Y0 or Y1 (Lemma 10.4a). Let us consider the worse case N0 = Y0 = H. Then we must have R0 = Y4 , Y7 or Y10 . If R0 = Y4 , then k(R0 W ) = 8 and k(Q) = 6. If R0 = Y7 , then k(R0 W ) = 10 and k(Q) = 5. Thus it remains to examine the case where R0 = Y10 . Here k(Q) = 10 and kg (N ) ≤ k(Q) · k(R0 ) = 20 by part (i) of Proposition 10.2a, and kg (N V ) ≤ k(Q) · k(R0 W ) = 140 by part (ii). Again we improve the latter bound by using part (iii) of this proposition. Let v = w1 + w2 be a 2-vector in V . Then w1 N = W1] and CN (w1 )/R2 is cyclic of order 4 (Lemma 10.4a). CN (w1 ) is faithful on W2] since it intersects R1 = CN (W2 ) trivially (being fixed point free on W1] ). Similarly CN (v) ∩ R2 = 1 and so |CN (v)| = 1, 2 or 4. As mentioned above CN (w1 ) (being of order 8) has at most 5 orbits on W2] . At any rate, kg (N V ) ≤ kg (N ) + (1 · 5) · 4 ≤ 40. The case N0 = Y1 is treated similarly. 2

n = 3: By Proposition 10.2a we may assume that 25|N0 | > 625 = |V | 3 . Thus |N0 | > 24 and so N0 = Y0 or Y1 by Lemma 10.4a, or N0 is a Sylow 2-subgroup of H = Y0 . Note that g permutes the Ti = CN (Wi ) cyclically (i = 1, 2, 3). Let T = T1 T2 , and let T0 be the image of T /T1 (∼ = T /T2 ) in N0 . Using that R3 = T1 ∩ T2 we see that T /R3 ∼ = T0 × T0 . From R = R1 × R2 × R3 we infer that the Ti are pairwise distinct, and from T1 R = T1 R1 and R1 ⊆ T2 we get T1 R ∩ T2 R = (T1 R ∩ T2 )R = (T1 ∩ T2 )R1 R = R.

192


Thus T /R ∼ = Q0 × Q0 , and Q = (N1 /R1 )∆N/T (N2 /R2 ) ∼ = Q0 ∆N0 /T0 Q0 is a fibre-product, identifying the Ri with their images in Ni = N/Ti . We have a group extension T0 /R0 ½ Q ³ Q0 . Thus k(Q) ≤ k(T0 /R0 ) · k(Q0 ) by (1.7b). We assert that kg (Q) ≤ |Q0 |. For any y ∈ N , the element gy does not centralize a nontrivial element in T1 R/R = T1 R1 /R. Hence CQ (gy) maps injectively into N/T1 R ∼ = N0 /R0 = Q0 , and by (1.7c) kg (Q) =

1 X |CQ (gy)| ≤ |Q0 |, |Q| y∈Q

as asserted. By Proposition 10.2a, kg (N V ) ≤ kg (Q) · k(R0 W ). Hence we may assume that |Q0 | ≥ 625/k(R0 W ) ≥ 25. This forces that N0 = H = Y0 and that R0 = Y10 . Thus k(Q0 ) = 10 and k(R0 W ) = 14. Further T0 is a normal subgroup of H, and by Proposition 10.2a we may assume that k(Q0 ) · k(T0 /R0 ) ≥ k(Q) > 625/k(R0 W ). It follows that k(T0 /R0 ) ≥ 5. This implies that T0 = Y0 , Y1 or Y3 (Lemma 10.4a). ] Now T1 /R3 ∼ = T0 has 1 or 3 orbits on W (depending on whether 2

T0 = Y0 , Y1 or Y3 ). Let v = w1 + w2 + w3 be a 3-vector in V (wi ∈ Wi] for i = 1, 2, 3). Then w1 N = W1] and CN (w1 )/T1 is cyclic of order 4, and CN (w1 ) has at most 3 orbits on W2] . From the structure of N0 /T0 we infer that CN (w1 )∩T2 properly contains R3 (which has 12 regular orbits on W3] ). It follows that CN (w1 +w2 ) = CN (w1 )∩CN (w2 ) has at most 8 orbits on W3] . Clearly CN (v)/CT1 (v) is isomorphic to a subgroup of CN (w1 )/T1 . Similarly, CT1 (v) ⊆ CT1 (w2 ) and so CT1 (v)/CR3 (v) is isomorphic to a subgroup of CT1 (w2 )/R3 . Since R3 is fixed point free on W3] and R3 ⊆ CN (w1 + w2 ), therefore CN (v) is a 2-group of order dividing 1·4·4 = 16, even |CN (v)| ≤ 8 unless T0 = H. In this latter case CN (w1 ) is transitive on W2] . At any rate, since (3 · 8) · 8 > (1 · 8) · 16, application of part (iii) of Proposition 10.2a yields that kg (N V ) ≤ kg (N ) + (1 · 3 · 8) · 8 ≤ 288. Here we use that kg (Q) ≤ |Q0 | = 48 and so kg (N ) ≤ kg (Q) · k(R0 ) ≤ 96, in view of part (i) of that proposition. ¤


193

Theorem 10.4d (Gluck–Magaard–Riese–Schmid). Let (G, V ) be induced from the nonreal reduced pair (H, W ) in characteristic p = 5. Then we have k(GV ) < |V |. Proof. We argue by induction on the degree n of (G, V ), the degree of the permutation group S = G/N . We may assume that n ≥ 3 (Lemma 10.4b). We may also assume that k(XV ) ≤ |V | for each proper subgroup X of G. For otherwise repeated application of (1.7b) yields that there exist a subnormal subgroup Y ¢of X and an irreducible Y -constituent U of V such ¡ that k (Y /CY (U )) · U > |U |. (Use that all modules are completely reducible. Start with an irreducible X-submodule U1 of V , let X1 = CX (U1 ), V1 = V /U1 and note that ¡ ¢ k(XV ) ≤ k (X/X1 )U1 · k(X1 V1 ). ¡ ¢ If k (X/X1 )U1 > |U1 | take U = U1 . Otherwise pick an irreducible X1 submodule U2 of V1 , and let X2 = CX1 (U2 ), V2 = V1 /U2 , etc. .) By Theorem 5.2b there is no real vector in U for Y . By Theorem 8.4, (Y /CY (U ), U ) is induced from a nonreal reduced pair (in characteristic 5). But (H, W ) is the unique, up to isomorphism, nonreal reduced pair in characteristic p = 5 by the classification of these pairs. Clearly the degree of (Y /CY ¢(U ), U ) is ¡ less than n. Thus by the inductive hypothesis k (Y /CY (U )) · U < |U |, a contradiction. Sn 2 By Proposition 10.4c for any g ∈ Gr i=1 Gi we have kg (N V ) ≤ |V | 3 . Let r be the number of conjugacy classes of S = G/N = (GV )/(N V ) which Sn are contained in i=1 Gi /N . We have k(S) > r ≥ 1 (Jordan). Application of Theorem 1.7d yields that 2

k(GV ) ≤ k(G1 V ) + (k(S) − r)|V | 3 . ¡ ¢ By (1.7b) k(G1 V ) ≤ k(H1 W1 ) · k CG (W1 ) · (W2 ⊕ · · · ⊕ Wn ) . Clearly |CG (W1 )| < |G|, hence k(CG (W1 ) · V ) ≤ |V |. ¡On the other hand CG (W¢1 ) · V = W1 × CG (W1 ) · (W2 ⊕ · · · ⊕ Wn ), and so k CG (W1 ) · (W2 ⊕ · · · ⊕ Wn ) ≤ |W |n−1 . It follows that k(G1 V ) ≤ k(HW ) · |W |n−1 = 20 · |W |n−1 . Since S √ n−1 by Proposition 9.4b. Consequently is a 50 -group and n ≥ 3, k(S) ≤ 3 √ n−1 2n − 1)|W | 3 = (0.8 + cn )|V | < |V | k(GV ) ≤ 20 · |W |n−1 + ( 3 for each n ≥ 3, because c3 = 0.08 and cn < 0.6n in general. We just use that |W | = 25. ¤

194


10.5. Summary Let G be a p 0 -subgroup of GL(V ) = GLm (p) for some prime p and some integer m ≥ 1 (|V | = pm ). With Theorem 10.4d the proof of the k(GV ) theorem is completed. Let us recall the basic steps: • We may assume that G is irreducible on V (Proposition 3.1a). • We may assume that there is no real vector in V for G (Theorem 5.2b). • The minimal counterexamples (G, V ), with respect to |V |, which admit no real vector, can be described via Clifford theory and lead to the so-called nonreal reduced pairs (Theorem 5.4). • The nonreal reduced pairs have been classified (Theorems 6.1 and 7.1). • If G is irreducible on V and if there is no real vector in V for G, then (G, V ) is induced from a nonreal reduced pair (H, W ), and G is not far from being a wreath product Hwr S for some permutation group S (Theorems 8.4 and 8.5c). • The cases where (G, V ) is nonreal induced, in the above sense, can be treated using upper bounds for the class numbers of permutation groups (Theorems 9.3 and 10.3, 10.4d). Theorem 10.5a. We have k(GV ) ≤ |V |, where equality can hold only if there is a strongly real vector in V for G. Proof. We have proved that k(GV ) ≤ |V | (see above). Suppose we have k(GV ) = |V |. Then k(Gi Vi ) = |Vi | for each irreducible submodule Vi of V and Gi = G/CG (Vi ) (Proposition 3.1a), and G is the direct product of the Gi . Combining Theorems 10.3, 10.4d and 8.4 we obtain that, for each i, there is a real vector vi ∈ Vi for Gi , which by Theorem 5.2b must be P ¤ strongly real. Now v = i vi is strongly real for G. Theorem 10.5b. We have k(G) ≤ |V | − 1 = pm − 1, and equality holds if and only if G is a Singer cycle in GLm (p). Proof. Note that k(G) < k(GV ) as G is a proper quotient group of GV = V : G. If k(G) = |V |−1, then G must be abelian and act transitively on V ] by Proposition 3.1b. Hence G is a Singer cycle in GL(V ). The converse is known from Sec. 3.2. ¤

Chapter 11

Possibilities for k(GV) = V It appears to be rather intricate to characterize the pairs (G, V ) where k(GV ) = |V | (in the usual coprime situation). By Proposition 3.1a it suffices to consider the case that G is irreducible on V . Assuming k(GV ) = |V | we shall establish certain congruences which indicate that, at least for large characteristics, G should be a Singer cycle in GL(V ). Unfortunately Clifford theory seems to fail in attacking this question. 11.1. Preliminaries Let p be a prime, G a finite p 0 -group, and let V be a finite faithful Fp Gmodule. Assume (without loss of generality) that G is irreducible on V . Proposition 11.1a. Suppose that k(GV ) = |V |. Then we have the following partial results: (i) If G has a regular orbit on V , then G is abelian. (ii) If G is transitive on V ] , then either G is a Singer cycle in GL(V ) or |V | = 23 and G is the Frobenius group of order 21, or |V | = 32 and G is semidihedral of order 16. (iii) If there is v ∈ V such that H = CG (v) is abelian, then k(HV ) = |V | and H acts on each irreducible H-submodule of [V, H] as a Singer cycle. (iv) There is a strongly real vector in V for G. Proof. This has been established in Theorems 1.5d, 3.4d and 10.5a.

¤

We know that k(GV ) = |V | if G is a Singer cycle, and also in the two further cases described in (ii) above. (For the dihedral subgroups G of GL2 (3) = GL(V ) the equality also holds.) Recall that Knörr’s generalized character δV of G is given by δV (x) = |V : CV (x)|; cf. Eq. (3.3b). Let e = p · exp (G), K = Q(e2πi/e ) and R be the ring of integers of K. Let Γ be the subgroup of Gal(K|Q) fixing each p0 -root of unity in K. So Γ ∼ = Gal(Q(εp )|Q) where εp = e2πi/p . Let p|p be a prime ideal of R above p. Then p0 = p ∩ Q(εp ) is the unique (totally ramified) prime ideal of R0 = Z[εp ] = R ∩ Q(εp ) above p. We have p0 = (1 − εp )R0 . 195

196


Lemma 11.1b. Suppose that k(GV ) = |V |. Let v ∈ V be strongly real for G, and let p 6= 3. Then H = CG (v) maps into SL(V ), and there is an integer-valued generalized character ψ of H such that ψ(1) = 1 and ψ 2 = δV (on H). We have hψχ, θiH = ±1 for all χ ∈ Irr(G) and θ ∈ Irr(H). Proof. If H does not map into SL(V ), clearly p is odd. Combining Theorems 5.2a and 3.3d yields that then k(GV ) ≤ p+3 2 |V |. But this is impossible by assumption. By Lemma 5.2a there exists ψ as asserted. From Theorem 3.3c it follows that hψχ, ψχiH = k(H) for all χ ∈ Irr(G). Let θ ∈ Irr(H). Then X |H|hψχ, θiH = ψ(h)χ(h)θ(h−1 ) ≡ χ(1)θ(1) (mod p), h∈H

because ψ(h)2 = |V : CV (h)| is a proper power of p for h ∈ H ] , as H is faithful on V , and ψ(1) = 1. On the other hand, χ(1)θ(1) is not divisible by p (Theorem 1.3b). Thus hψχ, θiH 6= 0, and the result follows. ¤ Keep the assumptions of the above lemma. As in Theorem 3.3d define the class function Ψ = Ψv on X = GV by letting Ψ(x) = |CV (h)|ψ(h) if x ∈ X is conjugate to hv for some h ∈ H, and letting Ψ(x) = 0 otherwise. Lemma 11.1c. For every χ ∈ Irr(X) the multiplicity hΨ, χi = ±1 if p = 2 or if V is in the kernel of χ, and otherwise it is a 2pth root of unity. Moreover, |H|hΨ, χi = χ(v −1 ) + pα for some α ∈ Z[εp ]. Proof. We make use of Theorem 3.3d, and its proof. By its very construction as an induced class function, Ψ is a Z[εp ]-linear combination of characters of X = GV . Let us write fχ = hΨ, χi for χ ∈ Irr(X). So fχ ∈ Z[εp ]. Using Frobenius reciprocity one gets X |H|fχ = ψ(h)χ(h−1 v −1 ) = χ(v −1 ) + pα h∈H

for some α ∈ R, because ψ(h) is divisible by p for all h ∈ H ] . Observe that χ(v −1 ) ∈ R0 = Z[εp ] and χ(v −1 ) ≡ χ(1) 6≡ 0 (mod p0 ) since p does not divide χ(1) (Theorem 1.3b). We see that α ∈ Q(εp ) ∩ R = R0 and, of course, that fχ 6= 0. As in Theorem 3.3d one also computes that hΨ, Ψi = P 1 2 h∈H |CV (h)|ψ(h) = |V |. |H|

Possibilities for k(GV)= V

197

Recall that for χ ∈ Irr(X), σ ∈ Γ and h ∈ H we have χσ (hv) = χ(hv)σ = χ(hv s ) if σ maps εp to εsp . It follows that fχσ = hΨ, χσ i 6= 0. Q Let N denote the norm for Q(εp )|Q. We know that N(fχ ) = σ∈Γ fχσ is a nonzero integer. By the arithmetic–geometric mean inequality (1.5b) 2 1 X σ2 |fχ | ≥ |N(fχ )| p−1 , p−1

σ∈Γ

and we have equality if and only if all |fχσ |, σ ∈ Γ, agree. Now by hypothesis k(X) = |V |, that is, |Irr(X)| = |V |. Observe that (p − 1)|V | = (p − 1)hΨ, Ψi =

X

X

|fχσ |2 .

χ∈Irr(X) σ∈Γ

Thus, for each χ ∈ Irr(X), we have equality above with |N(fχ )| = 1, whence |fχσ | = 1 for every σ ∈ Γ. Being a cyclotomic integer this implies that fχ is a root of unity (in R0 = Z[εp ]). Hence fχ = hΨ, χi = ±1 when p = 2, and otherwise it is a 2pth root of unity. If V is in the kernel of χ, then fχ = hΨ, χi =

1 X ψ(h)χ(h−1 ) = hψ, χiH = ±1 |H| h∈H

by the preceding lemma. This completes the proof.

11.2.

¤

Some Congruences

We keep the assumptions made in the preceding section. In particular V is an irreducible, faithful, coprime Fp G-module, X = GV and k(X) = |V |. The following congruences are established in [Schmid, 2005] and indicate that G should be a Singer cycle when p is large enough. For p = 2, 3 the congruences tell us nothing. For p = 2 the group G has odd order and so the Burnside congruence (1.5a) applies.

198


Theorem 11.2a. For any irreducible character χ of X = GV we have χ(1) ≡ ±1 (mod p), and k(G) ≡ |G| ≡ ±1 (mod p). Proof. We may assume that p ≥ 5. As above let v ∈ V be strongly real for G, and let H = CG (v). Let ψ and Ψ be the generalized characters of H and G studied in Lemmas 11.1b and 11.1c, respectively. Let χ ∈ Irr(X) be an irreducible character of X. Then hΨ, χi = ±1 if p = 2 or if V is in the kernel of χ, and hΨ, χi is a 2pth root of unity otherwise (11.1c). Letting R0 = Z[εp ] and p0 = (1 − εp )R0 be as before, we have hΨ, χi ≡ ±1 (mod p0 ), at any rate. We have also seen that |H|hΨ, χi = χ(v −1 ) + pα for some α ∈ R0 . We conclude that ±|H| ≡ |H|hΨ, χi ≡ χ(1) + pα ≡ χ(1) (mod p0 ). It follows that χ(1) ≡ ±|H| (mod p). Taking for χ the 1-character of X we get that |H| ≡ ±1 (mod p), and this in turn shows that χ(1) ≡ ±1 (mod p) in general. Picking χ = χv,1 as described in Proposition 3.1b, which is an irreducible character of X of degree |G : H|, we get |G : H| ≡ ±1 (mod p). Thus |G| = |H| · |G : H| ≡ ±1 (mod p). Finally |G| =

X

χ(1)2 ≡ k(G) (mod p),

χ∈Irr(G)

completing the proof.

¤

Theorem 11.2b. Let H = CG (v) for any vector v ∈ V . Then k(H) ≡ |H| ≡ ±1 (mod p) and θ(1) ≡ ±1 (mod p) for every θ ∈ Irr(H). Proof. Let χ = χv,1 in the notation of Proposition 3.1b. So χ is an irreducible character of X = GV with degree |G : H|. Thus |G : H| = χ(1) ≡ ±1 (mod p) by the preceding theorem. Similarly, for every θ ∈ Irr(H) the degree χv,ζ (1) = χ(1) · θ(1) ≡ ±1 (mod p) likewise. Hence θ(1) ≡ ±1 (mod p) and, therefore, |H| =

X

θ(1)2 ≡ k(H) (mod p).

θ∈Irr(H)

Use finally that |H| = |G|/|G : H|.

¤


11.3.

199

Reduced Pairs

Though Clifford reduction seems to fail in attacking the present problem, it is of some interest to examine the reduced pairs (G, V ). We use the notation introduced in Sec. 5.5. Hence V is a faithful F G-module where F = Fr for some power r of the prime p, the core E of G is absolutely irreducible on V , and G0 = NGL(V ) (E). Also Z = CG0 (E) ∼ = F ? , and χ is the Brauer character of G (and of G0 ) afforded by V . Proposition 11.3a. Suppose (G, V ) is a reduced pair of quasisimple type. Then k(GV ) < |V |. Proof. Assume k(GV ) = |V |. Then by Proposition 11.1a, (i) there is no regular G-orbit on V . We may appeal to Theorem 7.2a. The minimal point stabilizers H = CG0 (v) listed there are abelian in most cases, in which case part (iii) of Proposition 11.1a applies. If H is nonabelian, the desired contradiction follows from Theorem 11.2b. For instance, the case r = 7, H∼ = D12 is ruled out as there is ζ ∈ Irr(H) with ζ(1) = 2. Hence it remains to examine the permutation pairs (G, V ), where the core E is an alternating group Ad+1 and V is the deleted permutation module over F of dimension d, with p ≥ d + 2 (Example 5.1a). Since there is no regular vector in V for G, r = p = d + 2 or d + 3. At first, G can be any subgroup of G0 = S × Z containing E, where S ∼ = Sd+1 acts on V in the natural way. We know that there is a vector u ∈ V such that CG0 (u) is cyclic of order dividing p − 1 (Example 5.1a). From k(GV ) = |V | and Theorem 11.2b we infer that |CG (u)| = |CG0 (u)| = p − 1. This forces that p − 1 = d + 1. We next pick a strongly real vector v ∈ V for G0 as in Example 5.1a, Pd namely v = dw0 − i=1 wi (where {wi }di=0 is a permutation basis). Then 0 H0 = CG0 (v) ∼ = Sd and ResG H0 (V ) is the natural permutation module. Since the transpositions in H0 act with determinant −1 on V and since p ≥ 7, H = CG (v) ∼ = Ad by assumption and Lemma 11.1b. So |H| = 21 d! = 1 2 (p − 2)!, and from Theorem 11.2b we get that (p − 2)! ≡ ±2 (mod p). It follows that (p − 1)! ≡ ∓2 (mod p). On the other hand, (p − 1)! ≡ −1 (mod p). This forces that p = 3, a contradiction. ¤ Remark . Recently [Guralnick–Tiep, 2005] have shown that the inequality k(GV ) ≤ 21 |V | holds for each reduced pair (G, V ) of quasisimple type.

200


Proposition 11.3b. Suppose (G, V ) is reduced of extraspecial type. Then k(GV ) < |V | except possibly when r = 3. Proof. Assume that k(GV ) = |V | and that r 6= 3. Here E is a q-group of extraspecial type for some prime q 6= p, and |E/Z(E)| = q 2m for some integer m ≥ 1. By Proposition 11.1a, (i) there is no regular G-orbit on V . Similarly, there is no v ∈ V such that CG (v) 6= 1 is an elementary abelian 2-group. For then r = 3 by part (iii) of Proposition 11.1a, which has been excluded. There is also no strongly real vector v ∈ V such that H = CG (v) 6= 1 is cyclic. For then k(HV ) = |V |, and from Proposition 3.1a it follows that HV = H1 V1 × · · · × Hn Vn where either Hi = 1 (and |Vi | = r) or Hi 6= 1 is a Singer cycle on Vi . Since H is cyclic, the orders of the Hi are pairwise prime to each other. Since V is self-dual as an F H-module, this implies that each Vi is self-dual as an F Hi -module (and H = Hi for some i unless p = 2). From Lemma 4.1c it follows that |Hi | = 2 and |Vi | = 3 = r for some i, against our assumption. Probably the case r = 3 is not exceptional. But excluding r = 3 enables us to appeal to Theorem 6.3b. We get that q = 3, m ≤ 3 or q = 2, m ≤ 5, and r and p are suitably bounded above (Comments 6.3c). Let first p = 2. Then we make use of Theorem 6.4 (including its proof). We have E ∼ for m = 2 or 3 and GZ = X ◦ Z where X = E : H for = 31+2m + some point stabilizer H = CG (v), where v is strongly real for G. If m = 2 then H is cyclic of order 5, which is impossible. So we have m = 3, in which case H either is cylic of order 7 or is a Frobenius group of order 21. The former case cannot happen. From Theorem 6.4 we also know that the Weil character χ = ξ takes the value χ(h) = −1 on the elements h ∈ H of order 7, and χ(y) = 0, 3 or −9 for the elements y ∈ H of order 3. In addition χ(y) = 0 for the noncentral elements y of E. The dimensions of the eigenspaces on V of the noncentral elements of X of prime order are not greater than 12. Since 2|X|r14 = 2 · 21 · 37 r14 < r27 for r ≥ 4, there is a regular G-orbit on V , contradicting Proposition 11.1a. Let next p be odd. If q = 3 then p ≥ 5, and we make use of Theorem 6.5 (and its proof). If m = 1, then either there is v ∈ V such that CG (v) is an elementary abelian 2-group, or (G, V ) is nonreal reduced of type (33+ ) and k(GV ) ≤ 21 |V | by Proposition 10.1. If m = 2, then we find v as


201

before, or r = p = 7 in which case χ is an irreducible character of G of degree χ(1) = 32 6≡ ±1 (mod 7), contradicting Theorem 11.2a. Let m = 3. We must have r = 19, 13 or 7 (Comments 6.3c). Here χ is an irreducible character of G of degree χ(1) = 33 , so that Theorem 11.2a applies for r = p = 19. For r = 13 or 7 we have G 6= G0 , and G maps onto a proper ¯ of Sp6 (3). Checking the possible groups G ¯ and their irreducible subgroup G irreducible character degrees gives the result. Let q = 2 (and p 6= 2). There is no problem with m = 1 (Proposition 6.6a). For m = 2 we have r ≤ 83 (Comments 6.3c). If E ∼ = 21+4 + , by Proposition 6.6b we have p 6= 3, and there is a strongly real v ∈ V for G such that CG (v) ∼ = D12 when r = 7 and CG (v) ∼ = D8 otherwise (as point stabilizers cannot be cyclic or elementary abelian 2-groups). But then CG (v) has an irreducible character of degree 2, in contrast to Theorem 11.2b. For E ∼ either we get nonreal reduced pairs of type (25− ) or = 21+4 − r ≥ 11 and we find v as before. For p = 3, G maps into a 30 -subgroup ∼ ∼ 1+4 , then we find a of O− 4 (2) = S5 , and we get a regular orbit. If E = 20 strongly real v for G such that CG (v) is an elementary abelian 2-group for p = 3 and, otherwise, is isomorphic to S3 or A3 or to other explicit given groups. In all the p 6= 3 cases Theorems 11.2a and 11.2b apply. Let E ∼ = 21+6 . Here we have the (crude) bound r ≤ 139 (Comments −

6.3c). By Proposition 6.6c we have p 6= 3, and if r = 7 = p then there is v ∈ V such that CG (v) ∼ = S4 . Of course 24 6≡ ±1 (mod 7). For r ≥ 11 there is a strongly real vector v ∈ V for G such that H = CG (v) has a normal Sylow 3-subgroup S ∼ with index 2 or 4, and Theorem 11.2b applies. = 31+2 + 1+2m ∼ Let E = 2 with m = 3, 4, 5. For p = 3, by considering the +

irreducible 30 -subgroups of O+ 2m (2), one gets an elementary abelian 2-vector in V for G. Otherwise by Proposition 6.7a there is a strongly real vector v ∈ V such that CG0 (v) ∼ = 21+2m = GLm (2). Similar statements when E ∼ 0 (Proposition 6.7b). Apply Theorems 11.2a and 11.2b. Let finally E ∼ = 21+2m with m = 4, 5. As before one rules out the case −

p = 3. Otherwise by Proposition 6.7c there is a strongly real vector v ∈ V for G such that CG0 (v) ∼ = GLm (2), or m = 4 and CG0 (v) ∼ = G2 (2)0 . We have r ≤ 71 for m = 4 and r ≤ 23 for m = 5 (Comments 6.3c). Again Theorems 11.2a and 11.2b apply. ¤ Remark . The proof is less laborious once better upper bounds for r are available, and these can be obtained by computations with [GAP]. In our investigations in Chapter 6 we attempted to avoid computer calculations.

Chapter 12

Some Consequences for Block Theory There are various long-standing conjectures in modular representation theory. Presumably the most outstanding conjecture, and the most difficult one, is Brauer’s k(B) problem discussed in Chapter 2 of this monograph. In that chapter we already proved that the k(GV ) theorem implies Brauer’s k(B) conjecture for p-solvable groups. We shall recover this in terms of Brauer correspondence and blocks with normal defect groups. 12.1. Brauer Correspondence Let p be a prime, and let B be a p-block of the finite group G with defect group D. As usual k0 (B) denotes the number of (ordinary) irreducible characters belonging to B which are of height zero in the block. The Brauer correspondent b of B for NG (D) refers to Brauer’s first main theorem on blocks (mentioned in Sec. 2.5) and is that block of NG (D) with defect group D for which bG = B; sometimes b is called the germ of B (with respect to D). Observe that D = Op (NG (D)) by Theorem 2.3c. Let us recall some conjectures in modular representation theory. • Alperin–McKay conjecture: k0 (B) = k0 (b). • Olsson conjecture: k0 (B) ≤ |D : D0 |. • Brauer’s height zero conjecture: k0 (B) = k(B) if and only if D is abelian. • Broué conjecture: k(B) = k(b) if D is abelian. The Broué conjecture would follow from the Alperin–McKay conjecture and one-half of Brauer’s height conjecture. It would also be a consequence of Alperin’s weight conjecture [Alperin, 1987], or of the overall conjectures by [Dade, 1992]. From the k(GV ) theorem it follows that the above conjectures hold at least locally: Theorem 12.1a. We have k(b) ≤ |D| and k0 (b) ≤ |D : D0 |. Further k0 (b) = k(b) if and only if the defect group D is abelian. 202

Some Consequences for Block Theory

203

So if the Broué conjecture holds, Brauer’s k(B) problem would be settled for abelian defect groups. The Alperin–McKay conjecture is known to be true for p-solvable groups [Okayama–Wajima, 1980], and for some classes of finite groups G (including the symmetric groups and certain groups of Lie type). Brauer’s height conjecture has been also settled for p-solvable groups [Gluck–Wolf, 1984]. Hence we have the following. Corollary 12.1b. Suppose G is p-solvable. Then we have k(B) ≤ |D| and k0 (B) ≤ |D : D0 |. Also, k0 (B) = k(B) if and only if D is abelian. 12.2.

Clifford Theory of Blocks

It has been shown by [Reynolds, 1963] that Brauer’s height zero conjecture holds for blocks with normal defect groups. In order to prove Theorem 12.1a we have to appeal to his work. This is Clifford theory of blocks, dealing with root blocks. If B is a p-block of G with defect group D, there is a block b of H = DCG (D) with bG = B (Sec. 2.5). Each such b is a called a root of B in H. It follows from the first main theorem on blocks that these roots form a NG (D)-conjugacy class of blocks of H. Also, D is the unique defect group of such a root b by Theorem 2.3c. The inertial index of B is defined as the index of the inertia group in NG (D) of b (or its canonical character, see below) over H, and it is crucial that this is a p 0 -number. As in Chapter 2 we let Xp0 denote the set of p 0 -elements of a finite group X, and χp 0 = ResX Xp0 (χ) for a character χ of X. Lemma 12.2a. Suppose H = DCG (D) for some p-subgroup D of the finite group G, and let b be a p-block of H with defect group D. Then b contains a unique irreducible character θ having D in its kernel, and θp0 is the unique irreducible Brauer character in b. This canonical character θ belongs to a block of H/D with defect zero and is of height zero in b. We have k(b) = k(D); indeed Irr(b) consists of the characters θζ for ζ ∈ Irr(D), defined by θζ (y) = θ(y)ζ(yp ) if the p-part yp ∈ D and zero otherwise. For a proof we refer to [F, V.4.7]. Lemma 12.2b. Let D, H, b and θ be as above. Let T = I(θ) be the inertia group in NG (D) of θ. Then the block B = bG of G has defect group D if and only if |T : H| is not divisible by p.

204


Proof. By the first main theorem on blocks (and transitivity of block induction) we may assume that G = NG (D). Hence T = IG (θ). Since θp0 is the unique irrreducible Brauer character in b, T is the inertia group in G of the block b, that is, T is the stabilizer of the block idempotent eb to b (by conjugation). Thus the central character ωb of b is stable in G. We know that D ⊆ D0 for some defect group D0 of B (Sec. 2.5). By Lemma 2.3b there exists a conjugacy class c0 of G with c0 ⊆ H, ab (c0 ) 6= 0 and ωB (b c0 ) 6= 0 (in characteristic p), such that D0 is a defect group of c0 (see also Theorem 2.2b). Let x ∈ c0 with D0 ⊆ CG (x), and let c = xH . S Then c0 = i cti where the ti range over a right transversal of NG (c) in G. As in the proof for Theorem 2.3c, X 0 6= ωB (b c0 ) = ωb (b c0 ) = ωb (b c ti ) = |G : NG (c)|ωb (b c). i

Since NG (c) = CG (x)H and D0 is a Sylow p-subgroup of CG (x), it follows that p does not divide |G : D0 H|. Thus D0 ⊆ H if and only if |G : H| is not divisible by p, and then Lemma 2.3b implies that c and b have a defect group in common. ¤ Lemma 12.2c. Let N be a normal subgroup of the finite group X such that G = X/N is a p0 -group, and let Z = hεi be generated by a primitive exp (N )th root of unity ε. Suppose θ ∈ Irr(N ) is G-stable and θp0 ∈ IBr(N ). Then the Clifford obstructions µG (θ) = µG (θp0 ) agree in H2 (G, Zp 0 ). Proof. Let K = Q(ε) and R = Z(p) [ε], and let p be a prime of R above p and F = R/p. This R is a principal ideal domain. By Proposition 1.9a we may identify the Clifford obstruction µG (θ) = µKG (θ) with an element of H2 (G, Z) = H2 (G, Zp0 ) × H2 (G, Zp ). Since G is a p0 -group by hypothesis, H2 (G, Zp ) = 0. Clearly the Brauer character θp0 is G-invariant too. By Theorem 1.1d there is a KN -module W affording θ. Now ¡ ¢ M EndKG IndG Kτg N (W ) = g∈G

is a crossed product for some units τg sending W ⊗ g to W (τ1 = 1). By the preceding paragraph we may choose these units such that the factor set −1 τ (g, h) = τgh τg τh ∈ Zp0 for all g, h ∈ G. Let {tg }g∈G be a transversal to N in X, with t1 = 1, and let t(g, h) = t−1 gh tg th be the corresponding factor set. There are unique αg ∈ GL(W ) such that (w ⊗ tg )τg = (w)αg


205

for all w ∈ W , g ∈ G. For y ∈ N we have (wy)αg = (wy ⊗ tg )τg = (w ⊗ tg y)τg = (w ⊗ tg y tg )τg = (w ⊗ tg )τg y tg = (w)αg y tg (where we used that the tensor product is over KN and that τg is a KX-automorphism). −1 αg αh is the map w 7→ τ (g, h)wt(g, h) One similarly shows that α(g, h) = αgh on W . Let {wj } be a K-basis of W . Let U be the R-submodule of W generated by all (wj y)αg , y ∈ N , g ∈ G. Since U is finitely generated and torsion-free, it is a free R-module. Since U contains a basis of W and is contained in W , the rank of U is θ(1) = dim K W. From (w)αg y tg = (wy)τg for w ∈ W we infer that U is stable under N . From ¡ ¢ (w)αg τh = wα(g, h)αgh = τ (g, h)wt(g, h) αgh and τ (g, h) ∈ Zp0 ⊆ R? (unit group) we see that (U )αg ⊆ U for all g ∈ G. Since αg αg−1 = α(g, g −1 ) is the map u 7→ τ (g, g −1 )ut(g, g −1 ) on U , which is invertible, we may view the αg , via restriction, as elements of GL(U ) = GLR (U ). This in turn shows that U is a G-stable RN -lattice affording θ, yielding a projective R-representation of G with the same factor set τ as before. The F N -module V = U/pU affords θp0 . Identifying Zp0 with the group Zp0 (1 + p)/(1 + p) we get that µG (θp0 ) = µF G (θp0 ) agrees with µKG (θ) in H2 (G, Zp0 ), as desired. ¤ Proposition 12.2d (Reynolds). Suppose B is a p-block of G with normal defect group, D. Then there exists a group G0 having a normal Sylow psubgroup D0 ∼ = D and a p-block B0 of G0 such that the irreducible characters and Brauer characters of B0 and B are in 1-1 correspondence preserving heights, and B0 , B have the same decomposition and Cartan matrices. Proof. Let b be a root of B in H = DCG (D). By Theorem 2.3c, B is the unique block of G covering b. Let θ be the canonical character of b, and let T = IG (θ). Observe that D = Op (H) is the unique defect group of b and that T contains the inertia groups of all irreducible characters or Brauer characters in b. By Lemma 12.2b, p does not divide the inertial index |T : H| of B. We have B = bG = (bT )G , and it follows from Theorem 1.8b that the blocks B and bT behave (via character induction) as asserted for B, B0 (cf. Theorem 2.6b). Thus we may assume that T = G. Then b is the unique block of H covered by B. Let N = CG (D). Recall that the block idempotents to B and b are in the group algebra for N (Theorem 2.3c). Let bN be the block of N having

206


the same block idempotent as b. Then B is the unique block of G covering bN , and bN is the unique block of N covered by B. Note that Z(D) = N ∩D is the unique defect group of bN . ¯ = G/H, and let G(θ) ¯ ¯ is Let G be the representation group of θ. Since G 0 ¯ ¯ a p -group, G(θ) may be understood as a central extension of G with a cyclic ¯ group Zp 0 of order exp (N )p 0 (Lemma 12.2c). Let G(θ) = G(θ)∆ ¯ G be G the extended representation group. By Theorem 1.9c there is θb ∈ Irr(G(θ)) ∼ ¯ extending θ when viewed as a character of Ker(G(θ) ³ G(θ)) = N . Letting −1 ∼ e b θ be the unique linear constituent of θ on Ker(G(θ) ³ G) = Zp0 we have e given by ¯ a 1-1 correspondence χ ↔ ζ between Irr(G|θ) and Irr(G(θ))| θ), χ = θb ⊗ ζ (Theorem 1.9c). By Schur–Zassenhaus there is a complement Y¯ = Y /N to H/N in G/N . The group Y acts on D in the natural way (with kernel N ). Let b = D : Y and G e = D : Y¯ be the semidirect products. Then G b maps G e b onto G, and G maps onto G by replacing the direct product N × D by the b having the central product H = N ◦ D over Z(D). The characters of G “diagonal” of Z(D) × Z(D) (within N × D) in the kernel correspond to the characters of G. The group of the proposition is nothing but the extended representation group G0 = G(θ). This G0 has a normal Sylow p-subgroup D0 ∼ =D with CG0 (D0 ) = Z(D0 ) × Zp0 . By Theorem 2.3c there is a unique pe and this has defect group D0 . In particular block B0 of G0 covering θ, e By Lemma 12.2a, θ is of defect zero in N/Z(D), Irr(B0 ) = Irr(G0 |θ). so that θ(1)p = |N/Z(D)|p . Also, θp 0 is the unique irreducible Brauer character in bN , and Irr(bN ) consists of the characters θλ , for each (linear) λ ∈ Irr(Z(D)), satisfying θλ (y) = θ(y)λ(yp ) if yp ∈ Z(N ) and = 0 otherwise. We have (θλ )p 0 = θp 0 for each λ. Hence the above Clifford correspondence gives, by restriction to p 0 -elements, a bijection from IBr(B) e Note that (θ) b p0 is irreducible and (θ) e p0 = θ. e onto IBr(B0 ) = IBr(G0 |θ). Fix λ ∈ Irr(Z(D)). Regard θλ ∈ Irr(bN ) as a character of (N ×D)/D ∼ = N . The inertia group Tλ = IG b(λ) = IG b(θλ ) contains N × D. By Lemma 12.2c, µT¯λ (θλ ) is the restriction to T¯λ = Tλ /(N × D) of µG¯ (θ). Hence we ¯ may regard T¯λ (θλ ) as a subgroup of G(θ). Let Gλ and Sλ be the subgroups of G(θ) and G0 , respectively, mapping onto T¯λ , so that |G(θ) : Gλ | = |G0 : Sλ | = |G : Tλ |. By Lemma 12.2c we can further pick a character θbλ of Gλ extending θλ (in the usual sense) such that (θbλ )p 0 = ResGλ (θbp0 ). G(θ)


207

Hence on the p 0 -group Zp 0 , embedded into Gλ , θbλ is a multiple of θe−1 . By Theorem 1.9c we have a 1-1 correspondence ψ ↔ ϕ between Irr(Tλ |θλ ) and e given by ψ = θbλ ⊗ϕ. Here we use that each ψ ∈ Irr(Tλ |θλ ) lies Irr(G0 |λ−1 θ), b must be in the kernel of over λ and that the diagonal of Z(D) × Z(D) in G e ψ when inflated to a subgroup of this group. It follows that Sλ = IG0 (λ−1 θ) e viewed as a linear character of Z(D0 ) × Zp0 . is the inertia group of λ−1 θ,

G0 By Theorem 1.8b, χ = IndG Tλ (ψ) and ζ = IndSλ (ϕ) are irreducible. By Frobenius reciprocity χ ∈ Irr(B), as B is the unique block of G covering e Furthermore bN , and ζ ∈ Irr(B0 ) = Irr(G0 |θ). G b χp 0 = IndG Tλ (ψ)p 0 = IndTλ (ψp 0 ) = (θ)p 0 ⊗ ζp 0 .

The correspondence χ ↔ ζ between Irr(B) and Irr(B0 ) preserves heights, and B, B0 have the same decomposition and Cartan matrices. ¤ 12.3.

Blocks with Normal Defect Groups

In order to prove Theorem 12.1a we have to consider blocks with normal defect groups. The following lifting property for blocks has been established in [K¨ ulshammer, 1987]. Lemma 12.3a. Let D be a normal p-subgroup of G. Then the natural map G ³ G/D0 induces a bijection beteween the p-blocks of G and that of G/D0 . Proof. Let A = F G where F is a field of characteristic p. Let J be the kernel of the natural map F G ³ F [G/D0 ]. Thus J is the (left) ideal of A = F G generated by all t − 1, t ∈ D0 . But D0 is generated by all commutators [x, y] for x, y ∈ D, and ¡ [x, y] − 1 = x−1 y −1 (x − 1)(y − 1) − (y − 1)(x − 1) ∈ J(F D)2 ⊆ J(A)2 , and xy−1 = (x−1)(y−1)+(x−1)+(y−1) ≡ (x−1)+(y−1) (mod J(F D)2 ). Consequently J ⊆ J(A)2 . (Cf. [Jennings, 1941] for a more precise result; in fact one may replace D0 by the Frattini subgroup of D.) For any central idempotent e of A = F G, J + e is a central idempotent of A/J ∼ = F [G/D0 ]. If J + e = J + f for a central idempotent f of A, then e − ef = (e − 1)f is a central idempotent of A contained in J, hence is zero as J is nilpotent. Hence e = ef and, similarly, f = ef , hence e = f .

208


Now let e¯ be a central idempotent of A/J. By the usual lifting properties of idempotents there is an idempotent e of A such that e¯ = J + e [F, I.12.3]. It remains to show that e is a central idempotent of A. Since e¯ is central (and e¯(1 − e¯) = 0), J + eA(1 − e) = J + e(1 − e)A = J. Hence eA(1 − e) ⊆ J ⊆ J(A)2 and eA(1 − e) ⊆ J(A)2 (1 − e). Suppose we know already that eA(1 − e) ⊆ eJ(A)n (1 − e) for some integer n ≥ 2. Then eA(1 − e) ⊆ eJ(A)n−1 eJ(A)(1 − e) + eJ(A)n−1 (1 − e)J(A)(1 − e) ⊆ eJ(A)n+1 (1 − e). Since J(A) is nilpotent, this shows that eA(1 − e) = 0. Similarly (1 − e)Ae = 0. For any element a ∈ A we therefore have ea = eae + ea(1 − e) = eae and, analogously, ae = eae. Consequently ea = ae. Thus e is a central idempotent of A = F G, completing the proof. ¤ Theorem 12.3b. Suppose B is a p-block of G with normal defect group D. Then k(B) ≤ |D| and k0 (B) ≤ |D : D0 |. Proof. By Proposition 12.2d we may assume that D is a (normal) Sylow p-subgroup of G. It follows that G is p-solvable. Thus k(B) ≤ |D| by Theorems 2.6c/10.5a. By Lemma 12.3a there is a unique p-block B 0 of G/D0 corresponding to B. Clearly D/D0 is the unique defect group of B 0 . Let χ ∈ Irr(B 0 ). Then χ is an irreducible character of G (via inflation) with kernel Ker(χ) ⊇ D0 . Let θ be an irreducible (linear) constituent of ResG D (χ), and let T = IG (θ). By Theorem 1.8b, χ(1) = eχ |G : T |θ(1) = eχ |G : T | where the ramification index eχ is a divisor of |T /D|. Hence p does not divide χ(1). Therefore χ is of height zero. On the other hand, suppose χ ∈ Irr(B) is of height zero. Since the defect group D of B is a Sylow p-subgroup of G, χ(1) is not divisible by p. Let θ be an irreducible constituent of ResG D (χ). By Theorem 1.8b, θ(1) is a divisor of χ(1). Hence θ(1) is not divisible by p. Since D is a p-group, θ(1) is a power of p. Hence θ(1) = 1 and so χ has D0 in its kernel. By definition χ, as a character of G/D0 , belongs to B 0 . We have proved that k0 (B) = k0 (B 0 ) = k(B 0 ). From Theorem 10.5a it follows that k(B 0 ) ≤ |D/D0 |. ¤ Remark . The proof that k(B) = k0 (B) if and only if D is abelian [Reynolds, 1963] is quite similar.

Chapter 13

The Non-Coprime Situation As already mentioned in the preface, the inequality k(GV ) ≤ |V | does not hold if one drops the assumption of coprimeness (of |G| and |V |), in general. Examples 13.1. Let p be a prime and m ≥ 2 be an integer. (i) Suppose G is a Sylow p-subgroup of GLm (p), consisting of the ma(m) trices with ones in the main diagonal and zeros above it, and let V = Fp be the standard module. The structure of G is well understood [Huppert, 1967, III. §16]. One knows that G has (elementary) abelian normal sub2 2 groups of order pm /4 if m is even and of order p(m −1)/4 otherwise. In m [Higman, 1960] it is proved that k(G) < (m − 1)!pb 4 c . Following Hig2 1 man we show that k(G) > p 13 m . Since k(G) ≥ k(G/G0 ) = pm−1 , we may assume that m ≥ 12. For positive integers r ≥ s and r ≥ t, with 2r + s + t = m, consider the elements of G consisting of block matrices of the form   Is  A Ir   . J Ir B It Here A is a r ×s matrix having nonzero entries on the diagonal leading from the bottom right corner and zeros below this diagonal, J an r × r matrix with nonzero entries on its subsidiary diagonal and zeros otherwise, and B is a t×r matrix having nonzero entries on the diagonal leading from the top left-hand corner and zeros below this diagonal. Such (s, r, r, t) partitioned matrices are conjugate in G only if they are equal. Now consider those m matrices where r = d m 3 e and s = b 6 c. The number of freely disposable positions in matrices of the above form is cm = r(s + t) − s(s + 1)/2 − t(t + 2 1)/2 + 2, so that k(G) ≥ pcm . One verifies that cm > m 13 . (m)

(ii) G = GLm (p) acts irreducibly on V = Fp . We claim ¡ that the affine ¢ group GV = AGLm (p) has class number k(GV ) = k(G)+k AGLm−1 (p) > |V | (for m ≥ 2). It is known that pm − pm−1 < k(G) ≤ pm − 1 [Green, 1955]. For λ 6= 1V in Irr(V ) we have IG (λ) ∼ = AGLm−1 (p). Now argue by induction using¢ (1.10b) and the fact that G is transitive on Irr(V )] . Of ¡ course k AGL1 (p) = p. 209

210

The Solution of the k(GV) Problem (2m)

(iii) G = GL2 (p)wr Zm acts faithfully on V = Fp ¡ (2) and irreducibly ¢ (in the natural way). We have GV = Fp : GL2 (p) wr Zm and k(GV ) ≥

1 ³ p2 + p − 1 ´m |V | m p2 (2)

by Theorem 1.7a, because the affine group AGL2 (p) = Fp : GL2 (p) has class number k(GL2 (p)) + k(AGL1 (p)) = (p2 − 1) + p. (For p = 2 the group GV ∼ = S4 wr Zm has been already studied in Example 9.4c.) On the other hand, k(G) ≤ (p2 − 1)m + (p2 − 1)m − (p2 − 1) < p2m = |V | by Proposition 8.5d. Proposition 13.2 (Liebeck–Pyber). If G is an irreducible subgroup of GLm (p), then k(G) ≤ pcm for some constant c ≤ 10. This is Theorem 4 in [Liebeck–Pyber, 1997]. One might conjecture that here k(G) ≤ pm − 1, like in the coprime situation (Theorem 10.5b). The recent work by [Guralnick–Tiep, 2005] gives some evidence that this might be true at least when G is almost quasisimple. We turn to the computation of k(GV ). Inspection of the proof for Theorem 2.6c shows that if k(GV ) ≤ |V | holds in some more general situations, the k(B) conjecture holds for a corresponding wider class of p-constrained groups. First two general observations. Lemma 13.3. Suppose the finite group X has an abelian normal subgroup V . Let G = X/V , and let GV = V : G be the semidirect product (with G acting as before). Then k(X) ≤ k(GV ). Proof. This is immediate from formula (1.10b) since linear characters of V can be extended to their inertia groups provided these split over V . ¤ Lemma 13.4. Let V be a faithful Fp G-module, and let Op (G) = 1. Let Ve be the direct sum of the composition factors of V (in a fixed composition series). Then k(GV ) ≤ k(GVe ). Proof. Note that G is faithful on Ve since Op (G) = 1 (and since G is faithful on V ). Let {gj } be a set of representatives for the conjugacy classes of G. By Lemma 3.1c and (1.4a) X X X 1 |CV /[V,gj ] (h)|. k(GV ) = |C`(CG (gj )|V /[V, gj ])| = |CG (gj )| j j h∈CG (gj )

The Non-Coprime Situation

Hence k(GV ) =

1 |G|

P

P

g∈G

h∈CG (g)

211

|CV /[V,g] (h)|. There is a corresponding

identity for k(GVe ). Hence it suffices to show that dim CV/[V,g] (h) ≤ dim CV e/[V e,g] (h) for all elements g ∈ G and h ∈ CG (g). Let H = hg, hi. For any short exact sequence 0 → U → V → W → 0 of H-modules, U/[U, g] → V /[V, g] → W/[W, g] → 0 is an induced exact sequence of hhi-modules. This in turn gives rise to the exact sequence 0 → (W/[W, g])∗ → (V /[V, g])∗ → (U/[U, g])∗ for the dual hhi-modules, and then 0 → C(W/[W,g])∗ (h) → C(V /[V,g])∗ (h) → C(U/[U,g])∗ (h) is exact. By Theorem 1.4b the cyclic group hhi has on any module M the same number of fixed points as on its dual module M ∗ = Irr(M ). Consequently dim CV/[V,g] (h) ≤ dim CU/[U,g] (h) + dim CW/[W,g] (h). Now apply this (inductively) to the given G-composition series of V .

¤

Proposition 13.5 (Kovács–Robinson). Suppose the finite group X has a normal elementary abelian p-subgroup V of order pm . Assume G = X/V is faithful and completely reducible on V . If G is p-solvable, then there is a constant c, not depending on p or m, such that k(X) ≤ cm |V |. This has been proved in [Kovács–Robinson, 1993, Theorem 4.1]. Independence of the constant c from the prime p comes from the Fong–Swan theorem (mentioned in Sec. 2.7). In [Liebeck–Pyber, 1997] it has been shown that one actually may take c = 103. But by assumption and Theorem 2.3c, X has a unique p-block. Since the k(B) problem is solved for p-solvable groups, application of the Fong–Swan theorem, lifting the Brauer characters afforded by the irreducible summands of V to p-rational characters of G, and of a result by Schur [I, 14.19] therefore yields that 2

k(X) ≤ |X|p < pmp/(p−1) |V | in this case, and k(X) ≤ |V | if m < p − 1. Of course the constant cp = 2 pp/(p−1) depends on the prime p, but cp < 2 for p 6= 2, 3 and cp & 1 when p tends to infinity. One might conjecture that k(X) ≤ |V | in this p-solvable situation, provided p ≥ 5 is large enough. (By Example 13.1, (iii) this is false for p = 2, 3.) For arbitrary G a (weak) upper bound is given in [Liebeck–Pyber, 1997].

212


Proposition 13.6 (Robinson). Suppose the finite group X has a normal abelian subgroup V such that CX (V ) is a p-group. For any integer s ≥ 0 let ks (X) be the number of irreducible characters of X such that ps χ(1)p = P∞ |X|p . If k(X) ≤ |V |, then s=0 ks (X)/p2s ≤ 1/|V |, and equality only holds when V is a Sylow p-subgroup of X. Proof. Let |V | = pm and |X|p = pa . Then pm χ(1)p ≤ pa for all χ ∈ Irr(X) by Theorem 1.3b. Hence ks (X) = 0 for s < m, and ∞ X

ks (X)/p2s ≤ k(X)/p2m ≤ |V |/p2m = 1/|V |.

s=0

Pa 2s−2m Assume the equality is attained. Then = 1 and so s=m ks (X)p a m m ks (X) = 0 for s > m. Thus p = p χ(1)p = p , picking χ = 1X . ¤ In this proposition the group X again has a unique p-block, and ks (X) is the number of irreducible characters of X with height pa−s in this block (|X|p = p a ). This is motivated from [Brauer–Feit, 1959], counting characters with prescribed height (see also [F, V.§9]). Epilogue: There are lots of open questions in this area, and work continues. [Guralnick and Tiep, 2005] have studied thoroughly the case where G = X/V is almost quasisimple and faithful, irreducible on V . They proved that k(X) ≤ 21 |V | provided G does not involve An for 5 ≤ n ≤ 16 or a group of Lie type of (untwisted) rank at most 6 or a classical group with V related to the standard module. The exceptional cases have still to be examined. In a forthcoming paper D. Gluck has treated permutation pairs (Example 5.1a) in the non-coprime situation. As I understand, he is also investigating the question whether k(G) ≤ pm − 1 when G is an irreducible subgroup of GLm (p). Although the status of the classification of the finite simple groups is satisfactory [Aschbacher, 2004], a proof of the k(GV ) theorem independent of that would be greatly appreciated. Likewise challenging is Brauer’s k(B) problem. [Solomon, 2001] has the dream of returning in 100 years to ask about the meaning of the sporadic simple groups. My dream is to return in 100 years to ask, among others, about this mysterious Brauer problem.

Appendix A

Cohomology of Finite Groups Let G be a finite group and let V be a (right) G-module. The (Tate) cohomology is a Z-sequence of abelian groups Hn (G, V ) defined by H0 (G, V ) = P V G /V trG , where V G = CV (G) is the fixed module and trG = g∈G g in the group algebra ZG, which all vanish if V = IndG 1 (U ) for some abelian group U , together with connecting homomorphisms declared functorially for short exact sequences of G-modules. The existence is guaranteed via projective resolutions of the trivial G-module Z. For an introduction see [Hilton–Stammbach, 1971], which will be quoted as [HS]. In this book basically 1- and 2-cohomology is used, where we have the following familiar interpretations. Suppose V ½ X ³ G is an extension of the G-module V , so that X is a group with normal subgroup V and quotient group G = X/V , and X induces by conjugation the given Gmodule structure on V . Choosing a transversal {tg }g∈G to V in X we get a factor set (2-cocycle) τ ∈ Z 2 (G, V ) by defining τ (g, h) = t−1 gh tg th . This Z 2 (G, V ) is an (abelian) group by pointwise multiplication. Passing to other transversals gives a congruence relation on Z 2 (G, V ), and H2 (G, V ) is the corresponding quotient group. Conversely, each τ ∈ Z 2 (G, V ) defines such a group extension X = X(τ ), with underlying set G × V and multiplication (g, v) · (h, w) = (gh, vh + w + τ (g, h)). Thus H2 (G, U ) is in natural 1-1 correspondence with the (equivalence classes) of these group extensions, the zero element describing the semidirect product GV = V : G. In the semidirect product X = GV there may exist another subgroup Gα = {gvg | g ∈ G} complementing V in X (Gα ∩ V = 1 and X = Gα V ). Then α : g 7→ vg is a crossed homomorphism (1-cocycle) of G in V , and with respect to pointwise multiplication we get a group Z 1 (G, V ) acting regularly on the complements. Conjugacy of the complements in X (under V ) yields a congruence relation, and H1 (G, V ) is the corresponding quotient group. So H1 (G, V ) = 0 if and only if all complements are conjugate. Given any extension V ½ X ³ G of the G-module V , Z 1 (G, V ) may be identified with the group of all automorphisms of X centralizing V and G, and H1 (G, V ) accords with the quotient modulo those inner 213

214


automorphisms of X coming from V . If V = V G is a trivial G-module (X a central extension of G), then H1 (G, V ) = Hom(G, V ). Some general results on cohomology can be proved by describing them in cohomological dimension zero, say, and extending them functorially by dimension–shifting. is¢ due to the fact that there is a 1-induced G¡ This G module Vb = IndG Res (V ) having V as quotient and as submodule. For 1 1 n ≥ 1 the groups Hn (G, V ) = H−(n+1) (G, V ) are called homology groups. (A1) The exponent of Hn (G, V ) is a divisor of |G|. This is obvious for n = 0, hence is true for all n. For n = 1, 2 it leads, together with Sylow’s theorem (and Feit–Thompson), to the Schur–Zassenhaus theorem. (A2) Inflation, restriction: In dimension zero the restriction map ResG H for a subgroup H of G is defined by v + V trG 7→ v + V trH (v ∈ V G ), is induced from H ½ G → V on 1-cocycles, and is extended to all dimensions b by dimension–shifting. Observe that ResG H (V ) is a 1-induced H-module by Mackey decomposition. Let H be normal in G. Then V H is a G/Hmodule in the obvious way, and so is H0 (H, V ), and Hn (H, V ) for each n. The inflation map is v + V H trG/H 7→ v + V trG (v ∈ V G ) in dimension 0, is induced from G ³ G/H → V H ½ V on 1-cocycles, and is extended by dimension–shifting. There is a natural exact sequence Inf

Res

Inf

0 → H1 (G/H, V H ) → H1 (G, V ) → H1 (H, V )G/H → H2 (G/H, V H ) → K where K is the kernel of Res : H2 (G, V ) → H2 (H, V ) [HS, VI.81]. One can even describe the image of¡ H2 (G/H, V H ) in ¢ K as the kernel of a certain transgression map K → H1 G/H, H1 (H, V ) , e.g., using spectral sequences. (A3) Shapiro’s lemma: Let V = IndG H (U ) for some subgroup H of G and some H-module U . Then restriction to H and H-projection of V onto U gives rise to natural isomorphisms Hn (G, V ) ∼ = Hn (H, U ). This is easily verified for n = 0, hence holds everywhere (see also [HS], pp. 164 and 224). (A4) Restriction, corestriction: Let H be a subgroup of G. The corestricH tion map CorG → V G (with H is induced from the relative trace trH\G : V respect to any right transversal to H in G) in dimension zero, hence in general. Composition Cor◦Res is multiplication with |G : H| on Hn (G, V ) (see also [HS, VI.16.4]). It follows that if p is a prime not dividing |G : H|, then the restriction map Hn (G, V ) → Hn (H, V ) is injective on the p-components (which gives Gasch¨ utz’s splitting theorem).

Cohomology of Finite Groups

215

b = ZG is the regular module (group algebra) over Z, which maps (A5) Z onto Z with kernel IG , the so-called augmentation ideal, and contains 2 ∼ ZtrG ∼ = Z. One gets H1 (G, Z) ∼ = IG /IG = G/G0 [HS, VI .4.1]. The Schur multiplier of G is defined as M(G) = H2 (G, Z) or, equivalently, by the Hopf– Schur formula M(G) = R ∩ F 0 /[R, F] in any free presentation R ½ F ³ G of G [HS, VI.§9]. Suppose Z is any trivial G-module. Then, by the universal coefficient theorem [HS, VI.15.2], there is a natural exact sequence Inf

ρ

0 → Ext(G/G0 , Z) → H2 (G, Z) → Hom(M(G), Z) → 0. Here Ext(G/G0 , Z) governs the abelian extensions of G/G0 by Z, described by symmetric 2-cocycles, and ρ sends the cohomology class of a central extension Z ½ X ³ G to the (transgression) map M(G) → Z, which has the image Z ∩ X 0 . This may be derived on the basis of the Hopf–Schur formula. Using that Ext(G/G0 , Z) = 0 if Z is a divisible group we see that H2 (G, C? ) is the dual group to M(G). The central extension Z ½ X ³ G, with cohomology class x say, is called proper if Z ⊆ X 0 (ρx epimorphic), and a covering group (or Schur cover) if in addition M(G) ∼ = Z (via ρx ). If G = G0 is perfect, there b of G, up to isomorphism; every is a unique (universal) covering group G b automorphism of G can be lifted to G. (A6) The Schur multipliers of all quasisimple groups are known (Schur, Steinberg, Griess and others). In particular: M(An ) = Z2 for n 6= 6, 7 (M(A6 ) = M(A7 ) = Z6 ); M(GLm (2)) = 1 for m 6= 3, 4; M(Sp2m (q)) = 1 for all m and all odd prime powers q except when (m, q) = (1, 9); M(Sp2m (2)) = 1 for m ≥ 4; M(Ω± 2m (2)) = 1 for m 6= 2, 3 and with the exception M(Ω+ (2)) = Z × Z ; M(SUm (q)) = 1 for m ≥ 3 and 2 2 8 (m, q) 6= (4, 2), (4, 3), (6, 2). (A7) Module extensions: Let U ½ V ³ W be an extension of G-modules. We are only dealing with Z-split extensions. Even more, we shall be concerned mostly with F G-modules over a finite field F . These extensions are classified by ExtF G (W, U ) = H1 (G, HomF (W, U )). Here HomF (W, U ) ∼ = W ∗ ⊗F U as F G-modules (via diagonal actions), and W ∗ = HomF (W, F ) is the dual module to W . From projective resolutions of the trivial F G-module F it is immediate that only irreducible modules belonging to the principal block can have nontrivial cohomology. If γ is a group or field automorphism and V γ is the twisted F G-module thus obtained, Hn (G, V γ ) ∼ = Hn (G, V ) via γ. (All Ext-groups vanish if char(F ) - |G|; Maschke’s theorem.)

216


Remark . Let V be an F G-module and T 2 (V ) = V ⊗F V . Identify Λ2 (V ) = Alt2 (V ) with the alternating square, and let Sym2 (V ) be the symmetric square (fixed space of v ⊗ w 7→ w ⊗ v). Let G = SL(V ). If char(F ) 6= 2, then both Λ2 (V ) and Sym2 (V ) are irreducible F G-modules and T 2 (V ) = Λ2 (V ) ⊕ Sym2 (V ), as in (1.1). Otherwise Λ2 (V ) is a submodule of Sym2 (V ) with quotient module isomorphic to the twist V γ of V under the Frobenius automorphism γ of F , and T 2 (V ) is a uniserial F G-module with composition factors Λ2 (V ), V γ , Λ2 (V ). (2m)

(A8) Let V = Fp for some prime p and some m ≥ 1. Then M(V ) = Λ2 (V ) by the K¨ unneth theorem [HS, V.3.1]. Letting R ½ F ³ V be a free presentation of (minimal) rank 2m, we have a natural exact sequence M(V ) ½ R/[R, F]Rp ³ V . This sequence splits, and splits naturally through the pth power map when p is odd. The dual of this sequence agrees with the universal coefficient sequence V ∗ ½ H2 (V, Fp ) ³ Λ2 (V )∗ . Now regard V ∼ = V ∗ as the standard module for G = Sp2m (p). Since G is absolutely irreducible on V , HomG (V ⊗ V ∗ , Fp ) = EndG (V ) = Fp . In the odd case H2 (V, Fp ) ∼ = V ∗ ⊕ Λ2 (V )∗ as a G-module, and there is a unique alternating form 6= 0, up to scalar multiples, which is G-invariant. This defines the extraspecial group p1+2m (since the group of exponent p2 does + not admit the symplectic group G). For p = 2 we have a similar result 1+2m replacing the symplectic group by O± . 2m (2), defining thus the groups 2± (A9) The cohomology of certain classical groups on their standard modules and related modules is known [Griess, 1973], [Sah, 1977], [Bell, 1978]. It vanishes by (A2) whenever a nontrivial scalar multiplication appears. Also, (m) H1 (SLm (q), Fq ) = 0 for (m, q) 6= (2, 2r ) with r > 1 and (m, q) 6= (3, 2), (m) 2 H (SLm (q), Fq ) = 0 for (m, q) 6= (2, 2r ) with r > 2, (m, q) 6= (3, 3r ) with (2m) r > 1, and (m, q) 6= (3, 2), (4, 2), (5, 2), (3, 5); Hn (Sp2m (2), F2 ) = Z2 = (4) (2m) Hn (Sp2m (2) × Z2 , F2 ) for m > 1 and n = 1, 2, but H2 (Sp4 (2)0 , F2 ) = 0; (2m) further H1 (O± ) = 0 except for m = 3 and positive type, and 2m (2), F2 (2m) ± 2 H (O2m (2), F2 ) = Z2 for m > 2 and the same result for Ω± 2m (2) except for Ω+ (2) where the 2-cohomology vanishes (as it does for m = 1, 2). – The 6 computer was needed to compute the 2-cohomology of some orthogonal − groups on the standard module, e.g., for Ω+ 10 (2) and Ω12 (2) (Derek Holt). (A10) The nonzero element, if any, of the 2-cohomology of the orthog± onal groups O± 2m (2), Ω2m (2) and the groups Sp2m (2), Sp2m (2) × Z2 , on the standard modules is represented by the automorphism group of the corresponding 2-groups of extraspecial type [Griess, 1973].

Appendix B

Some Parabolic Subgroups The nonabelian finite simple groups fall into the following four classes: The alternating groups An (n ≥ 5), the finite classical groups, the exceptional groups of Lie type, and the 26 sporadic groups. In this appendix we are concerned only (for convenience) with the classical groups Sp2m (q) and − SO+ 2m (q), SO2m (q) (q a prime power, m ≥ 1). In Lie terminology the symplectic group is the simply connected group of type Cm whereas the orthogonal groups under consideration may be identified as those groups of type Dm (plus type) and 2 Dm (negative type) which are neither simply connected nor adjoint. In the orthogonal case there is a subgroup Ω± 2m (q) ± of index 2 in SO2m (q), which is the kernel of the Dickson invariant when q is even, and this is the unique subgroup with index 2 unless we have plus type and m = 2 = q. Let U be the standard module for these classical groups, and write G = G(U ) for the related isometry group (introduced above; we write G(U ) = O± 2m (q) in the even orthogonal case). Let F = Fq . Let τ denote either the symplectic form on U or the symmetric form to the quadratic form Q in the orthogonal case. A nondegenerate symplectic form requires even dimension, and then the form on U is uniquely determined up to isomorphism (and so is the group). In the orthogonal case Q : U → F satisfies Q(cu) = c2 Q(v) for c ∈ F , u ∈ U , and τ (u, v) = Q(u + v) − Q(u) − Q(v) defines the associated (symmetric) bilinear form on U . The form Q is nondegenerate if τ is nondegenerate. If q is odd, the theory of quadratic forms is that of symmetric bilinear forms. If q is even then τ necessarily is symplectic. Actually we are interested in quadratic forms only in characteristic 2, hence the assumption that dim F U = 2m is even. There are up to isomorphism two distinct nondegenerate quadratic forms Q = Q+ 2m (Witt index m) and Q = Q− on V (index m − 1). This means that U possesses maximal 2m totally singular subspaces of dimension m resp. of dimension m − 1; in the symplectic case we have index m. Reference for all this is [Dieudonné, 1971]. 217

218


(B1) Suppose U has index m. Then there exist totally singular subspaces W, W ∗ of dimension m such that U = W ⊕ W ∗ . Let P = NG (W ) and let L = NG (W, W ∗ ) be the normalizer of both subspaces, R = CG (W, U/W ) be the centralizer of both W and U/W . Then P = R : L where L ∼ = GL(W ) and W is the standard module of L and W ∗ is the dual module. Also R∼ = Alt2 (W ) as module for L in the orthogonal case, and R ∼ = Sym2 (W ) in the symplectic case. Proof. By hypothesis there is basis {e1 , · · ·, em } ∪ {e∗1 , · · ·, e∗m } of U such that τ (ei , e∗i ) = 1 for all i and all other scalar products of the ei , e∗j vanish, and that in adddition Q(ei ) = 0 = Q(e∗j ) in the orthogonal case. Then the hei , e∗j i are hyperbolic planes in U , and W = he1 , · · ·, em i and W ∗ = he∗1 , · · ·, e∗m i are maximal totally singular subspaces. µ ¶ x 0 ∗ Let x ∈ GL(W ) and y ∈ GL(W ). Then is an isometry of 0 y ∗ ∗ the symplectic space U if and only if τ (ei x, ej y) = τ (ei , ej ) = δij for all i, j. This means that y = (x−1 )t has µ to be the ¶ inverse transpose of x. In the x 0 orthogonal case, observe that preserves W and W ∗ and that 0 x−t Q(w + w∗ ) = 0 for w ∈ W and w∗ ∈ W ∗ if and only if τ (w, w∗ ) = 0. So µ

x 0 L={ 0 x−t

¶ | x ∈ GL(W )}

may be viewed as a subgroup of G in each case, and it is the stabilizer in G of W and W ∗ . Of course, L ∼ = GL(W ) and W is the standard module, W ∗ its dual. It is clear that L ∩ R = 1. Let x ∈ P . We find y ∈ L such that xy centralizes W . But U/W ∼ = W ∗ and so xy centralizes U/W as well, that is, xy ∈ R. Thus P = R : L. By (A7) it is clear that R is an F L- submodule of HomF (U/W, W ) ∼ = HomF (W ∗ , W ) ∼ = T 2 (W ). Let α : U/W → W be a linear map. View α as an endomorphism of U via P inflation. Write e∗i α = j aij ej . We have to examine when 1 + α is in G. In the symplectic case this requires that X X 0 = τ (e∗i , e∗k ) = τ (e∗i + aij ej , e∗k + akj ej ) = aki τ (e∗i , ei ) + aik τ (ek , e∗k ) j

j

Some Parabolic Subgroups

219

for all i, k. Thus the matrix (aik ) has to be symmetric. It is well known that T 2 (W ) is naturally isomorphic to Mm (F ) (sending ei ⊗ek to the elementary matrix eik ), in which case Sym2 (W ) maps onto the symmetric matrices and Alt2 (W ) onto the skew symmetric matrices. In the orthogonal case we compute 0 = Q(e∗i ) = Q(e∗i +

X j

aij ej ) = τ (e∗i ,

X

aij ej ) = aii

j

¡ ¢ P and, for i 6= k, 0 = Q(e∗i + e∗k ) = Q e∗i + e∗k + j (aij + akj )ej = aik + aki . Thus in this case the (necessary and sufficient) condition is that the matrix (aik ) is skew symmetric. The proof is complete. ¤ (B2) Suppose W 6= 0 is a totally singular subspace of U , and let P = NG (W ). There exists a totally singular subspace W ∗ of U such that U 0 = W ⊕ W ∗ is nondegenerate. Let U 00 = (U 0 )⊥ , L = NG (W, W ∗ , U 00 ) and R = CG (W, W ⊥ /W, U/W ⊥ ). Then P = R : L and L = Lw × G(U 00 ), where Lw ∼ = GL(W ) acts naturally on W and W ∗ is the dual F Lw -module, and where G(U 00 ) centralizes Rw = CR (W ⊥ ). Moreover, Rw ⊆ Z(R), R/Rw ∼ = W ⊗F U 00 as an L = Lw × G(U 00 )-module, and Rw ∼ = Sym2 (W ) 2 as an Lw -module in the symplectic case and Rw ∼ = Alt (W ) otherwise. Proof. By Witt’s theorem we may assume that W has the basis {e1 , ···, en } taken from the standard basis above (n ≤ m). Let then W ∗ = he∗1 , · · ·, e∗n i. So (B1) applies to U 0 = W ⊕ W ∗ , and U = U 0 ⊥U 00 is an orthogonal sum of nondegenerate spaces. Hence L = Lw × GL(U 00 ) is as asserted. ∼ W ∗ (through It is obvious that L ∩ R = 1. As a P -module U/W ⊥ = the given nondegenerate pairing). Thus to x ∈ P we find y ∈ Lw such that xy centralizes W and U/W ⊥ . We have W ⊥ = W ⊕U 00 . Hence each element in W ⊥ /W can be written as W + u00 for some unique u00 ∈ U 00 . We get a linear map α : U 00 → U 00 satisfying (W + u00 )α = W + u00 xy. It is easy to see that α ∈ G(U 00 ). This proves that P = R : L. The quotient R/Rw acts faithfully on W ⊥ , centralizing both W and W /W ∼ = U 00 , whence may be identified with HomF (U 00 , W ). In fact, for each linear map ϕ : W ⊥ /W → W consider its inflation to W ⊥ , and then 1 + ϕ gets an isometry of W ⊥ , which extends to an element of G by Witt’s theorem. Thus R/Rw ∼ = HomF (U 00 , W ) as an L = G(U 00 ) × Lw -module. 00 Now use that U is a self-dual G(U 00 )-module. ⊥

220


Since Rw centralizes W ⊥ ∼ = U/W and R centralizes U/W ⊥ and W , we have [U, R, Rw ] = 1 = [Rw , U, R]. It follows that [R, Rw , U ] = 1 (by the 3-subgroups lemma) and so Rw ⊆ Z(R). Now Rw centralizes U 00 and acts on the nondegenerate space U 0 = W ⊕ W ∗ and centralizes W , with U 0 /W ∼ ¤ = W ∗ . Apply (B1) to U 0 . Remark . From (A7) we infer that Lw acts irreducibly on Rw unless p = 2 and G(U ) = Sp2m (q) is a symplectic group (in which case Rw is indecomposable with 2 composition factors). Also, R/Rw ∼ = W ⊗F U 00 is an + 00 00 ∼ irreducible Lw × G(U )-module unless G(U ) = O2 (2) or SO+ 2 (q) for odd q, where G(U 00 ) is not irreducible on the standard module. (B3) Let W, P, R, L be as in (B2). Then P is a maximal parabolic subgroup in G = G(U ), with unipotent radical R and Levi complement L, except when G(U ) ∼ = SO+ 2m (q) and dim F W = m, in which case P is a maximal parabolic in Ω+ (q). The maximal parabolics are all of of this form, and 2m are determined up to conjugacy by the dimension of W (which may vary from 1 to the Witt index). Proof. It follows from the construction, and from (B2), that P = NG (R) is a proper subgroup of G and that R = Op (P ). The order of P is determined by n = dim F W. Indeed dim F U00 = 2m − 2n and so |R| = |R/Rw | · |Rw | = q n(2m−2n)+n(n±1)/2 , where the positive sign holds in the symplectic case. One similarly computes |L| = |GL(W )| · |G(U 00 )|. This shows that P contains a Sylow p-subgroup of G, even a p-Sylow normalizer, except when n = m and G ∼ = SO+ 2m (q). + So P is a parabolic in G or, in the exceptional case, in Ω2m (q) (where there are two classes which fuse in O+ 2m (q)). If P were not maximal in G (resp. in Ω+ 2m (q)), there were a maximal parabolic P0 = R0 : L0 properly containing P . Then at any rate R0 ⊂ R and U0 = CU (R0 ) ⊇ W . We even have W ⊆ U0 ∩ U0⊥ = U1 , and U1 is P0 -invariant. Either U1 is totally singular or q is even, U is orthogonal, and the subspace U10 of U1 of singular vectors has codimension 1. But then U10 is P0 -invariant and U10 ⊇ W . So P0 stabilizes a totally singular subspace W0 ⊇ W of U . Since P0 is a maximal subgroup, P0 is the normalizer of W0 , and W0 6= W as P0 6= P . We get a contradiction by comparing the orders of R and R0 . Witt’s theorem, and a survey of the Dynkin diagram, give the final uniqueness statement. ¤

Appendix C

Weil Characters Let G be any of the classical groups SLm (q), Sp2m (q) or SUm (q) for some m ≥ 1 and some power q = pf of a prime p, and let U be the natural G-module over Fq (resp. Fq2 in the unitary case). In the symplectic case assume that q is odd. Let Gu be the associated universal group, that is, Gu = GLm (q), CSp2m (q) or GUm (q), respectively. Then Gu /Z(Gu ) is the adjoint group, which may be identified with the normal subgroup of Aut(G) generated by the inner and diagonal automorphisms. Stimulated by [Weil, 1964] one associates to Gu and G certain distinguished characters related to extraspecial groups (or Heisenberg groups). It turns out that, with few exceptions, these Weil characters are faithful of minimal degree. If Z is a trivial (additive) U -module, every biadditive map τ : U ×U → Z is a factor set in Z 2 (U, Z), and if Z ½ E ³ U is an extension to τ , the commutator map on E = E(τ ) induces the alternating (symplectic) form given by [u, u0 ] = τ (u, u0 ) − τ (u0 , u). (C1) Let G = SLm (q) and Gu = GLm (q) for some integer m ≥ 2. The Pq−2 permutation character of Gu on U decomposes as πU = 2 · 1Gu + j=0 ξj , where the ξj are pairwise distinct (complex) irreducible characters of Gu satisfying ξ0 (1) = (q m −q)/(q−1) and ξj (1) = (q m −1)/(q−1) for j > 0. For m ≥ 3, and excluding (m, q) = (3, 2), (3, 4), (4, 2), (4, 3), these characters remain irreducible on G, and we have R0 (Lm (q)) = (q m − q)/(q − 1). Each faithful projective irreducible character of Lm (q) = PSLm (q) of degree m −1 then is such a Weil character ξj . at most qq−1 Proof. Let U ∗ be the dual Fq Gu -module to U , and let W = U ⊕ U ∗ . Let further Fq ½ E ³ W be the associated Heisenberg group. Thus E = E(τ ) is the set of pairs (w, z) in W × Fq with (w, z) · (w0 , z 0 ) = (w + w0 , z + z 0 + τ (w, w0 )) where the factor set τ : W × W → Fq is Fq -bilinear, given by τ ((u, λ), (u0 , λ0 )) = λ(u0 ). For g ∈ Gu we define (u, λ, z)g = (ug , λg∗ , z) where g ∗ is the inverse transpose of g. It is immediate that in this manner 221

222


Gu preserves the factor set τ and so Gu acts on E (faithfully) centralizing Z(E) ∼ = Fq . Even more, the subgroup A = U × {0} × {0} of E is Gu invariant and is the standard module, and A∗ = {0} × U ∗ × {0} is the dual module. Every nontrivial linear character λ of Z(E) ∼ = Fq gives rise to an irreducible character χ0 of E by inducing up to E the character 1A∗ × λ of A∗ × Z(E). Inducing up to E : Gu the linear character 1A∗ :Gu × λ of (A∗ : Gu ) × Z(E) we obtain an irreducible character χ of E : Gu extending χ0 . Let ξ be the restriction to Gu of this irreducible character. Using that A is a set of right coset representatives for (A∗ : Gu ) × Z(E) in E : Gu , with Gu acting transitively on A] , and using Mackey decomposition, we see that ξ = π is the permutation character of Gu on the set A ∼ = U . Of course ] Gu is transitive on U . Z(Gu ) ∼ = F? has (q m − 1)/(q − 1) regular orbits (cycles) on U ] . Hence q

we have q − 1 distinct constituents ξj0 of πU ] lying over the distinct linear characters λj of Z(Gu ), and all these ξj0 have degree (q m − 1)/(q − 1). The character ξ00 lying above the 1-character of Z(Gu ) is of the form ξ00 = 1Gu + ξ0 with ξ0 (1) = (q m − q)/(q − 1). Let ξj0 = ξj for j > 1. The Gu -sets U , U ∗ are isomorphic through the inverse transpose automorphism. Thus W ∼ = U × U as Gu -sets and ξ 2 = πW . Now Gu has exactly q + 3 orbits on the set U × U (being transitive on the set of linear independent pairs of vectors (u, u0 ) and on the sets (u, tu) for each t ∈ F?q ). Pq−1 From ξ = 2 · 1Gu + i=1 ξi and hξ, ξi = hξ 2 , 1Gu i = q + 3 we conclude that all ξi are irreducible and pairwise distinct. The restrictions to G = SL(W ) = SLm (q) of these characters need not be irreducible (nor distinct) for small m. We refer to [Tiep–Zalesskii, 1996, Theorem 3.1] for the final statements in (C1). One has R0 (Lm (q)) = 3, 6, 7, 26 for (m, q) = (3, 2), (3, 4), (4, 2), (4, 3), respectively. ¤ (C2) Let G = Sp2m (q) and Gu = CSp2m (q) for m ≥ 2 and odd q. The permutation character of G on U is of the form πU = ξ · ξ¯ = ξ ¦ · ξ ¦ for Gu -conjugate “generic” Weil characters ξ 6= ξ ¦ . We have ξ = ξ1 + ξ2 and ξ ¦ = ξ1¦ + ξ2¦ with irreducible (Weil) characters ξ1 , ξ1¦ and ξ2 , ξ2¦ of degree (q m − 1)/2 and (q m + 1)/2, respectively. Moreover, R0 (S2m (q)) = (q m − 1)/2, and every faithful projective irreducible character of S2m (q) = PSp2m (q) of 1 (q m − 1)(q m − q) is one of ξ1 , ξ2 , ξ1¦ , ξ2¦ . degree less than 2(q+1)

Weil Characters

223

The final statement is a result due to [Tiep–Zalesskii, 1996, Theorem 5.2]. The other assertions follow from Theorems 4.3c and 4.5a. Recall, in particular, that ξ and ξ ¦ are obtained by restriction to Sp2m (q) of the faithful irreducible characters of degree q m of the symplectic holomorph m and where U = E/Z(E) may be identified E : Sp2m (q), where E = p1+2f + with the standard module. (C3) Let G = SUm (q) and Gu = GUm (q), with m ≥ 3. For convenience exclude (m, q) = (3, 2), (4, 2), (6, 2) and (4, 3). Let X be the (standard) m m holomorph of E = p1+2f for odd p and of E = 21+2f otherwise, with + 0 ∼ X/E = Sp2m (q) (say). There is a distinguished subgroup Y of X such that Y ∩ E = Z(E) and such that Y E/E ∼ = Gu .2 is a field extension subgroup of Sp2m (q). We have a unique decomposition Y = Y0 × Z(E) for odd q, where we write Y0 = Gu .2 ⊇ Gu , and where Y 0 = Gu otherwise. In both cases U = E/Z(E) is the standard module for Gu , and the “generic” Weil character of Gu is defined as ξu = ResX Gu (χ) · µ, where χ is any faithful irreducible character of X of degree q m and where µ is the linear character of Gu of order gcd(q + 1, 2). This ξu is welldefined, rational-valued and satisfies ξu2 = πU . Fixing a linear character λu of Z(Gu ) of order q + 1 (which exists) let ξj be the constituent of ξu lying above λju (0 ≤ j ≤ q). Then ξ0 (1) = (q m + (−1)m q)/(q + 1) and ξj (1) = ξ0 (1) − (−1)m for j > 0, and the ξj are pairwise distinct irreducible (Weil) characters of Gu and remain so when restricted to G. We have ½ m (q − q)/(q + 1) if m is odd R0 (Um (q)) = (q m − 1)/(q + 1) otherwise, and each faithful projective irreducible character of Um (q) = PSUm (q) of m m +1 +q for odd m, and qq+1 otherwise, is one of the ξj . degree at most qq+1 Proof. For existence (and uniqueness) of Gu .2 in Sp2m (q) see [Aschbacher, 1984]; in the odd case it is an (almost) maximal subgroup, and determined up to conjugacy. Recall that Gu /G ∼ = Z(Gu ) are cyclic of order q + 1, inverted by Gu .2 since the outer field automorphism of order 2 sends any z to z q = z −1 . Let Zu = Z(Gu ) = hzu i, which acts on U as a group of scalar multiplications (with norm zu · zuq = 1). By assumption and (A6) the group G = SUm (q) is perfect and M(G) = 1. Using (A2), (A5) one shows that also M(Gu ) = 1 = M(Gu .2), and that Hn (Gu .2, U ) = 0 for all n.

224


Let α be an automorphism of X permuting the faithful linear characters of Z(X) = Z(E) transitively (Theorems 4.3c and 4.3d). By the cohomological triviality stated above, we find Y in X as claimed, and αinvariant. In the even case use that α is trivial on X/E. If |Z(E)| = p is odd, p does not divide |Gu .2/G| = 2(q + 1). In the even case Gu .2/G is dihedral (and q + 1 odd). Thus Y = Gu .2 × Z(E) when q is odd, and Y 0 = Gu otherwise (A5), where Gu is α-invariant and α gets inner on Gu in each case. So the faithful irreducible characters of X of degree q m agree on Gu , defining ξu , and ξu is rational-valued. As in Theorem 4.5a each element of Gu is good for U (use zu ). Hence ξu2 = πU by Theorem 4.4. Let ε = e2πi/(q+1) , and choose λu such that λu (zu ) = ε. Now Zu has on U ] just (q m − 1)/(q + 1) regular orbits. It follows from Theorems 4.4 and Pq 1.6c that ξu (zu ) = (−1)m (see also [I, 13.32]). Let ρ = j=0 λju denote the q m −1 2 0 u regular character of Zu . We know that ResG Zu (ξu ) = λu + q+1 · ρ. Hence

m u and ρ(zu ) = 0, ResG Zu (ξ) is not a multiple of ρ. Using that ξu (zu ) = (−1) and using elementary properties of sums of (q + 1)th roots of unity we get ½ 0 if m is even, a·ρ+λ u Puq ResG (ξ ) = Zu u a · ρ + j=1 λju otherwise

for some nonnegative integer a. Of course a = m

q m −1 q+1

when m is even,

−q otherwise. Let ξj denote the constituent of ξu lying over and a = qq+1 the linear character λju of Zu . Then ξ0 (1) = a + 1 when m is even, and ξ0 (1) = a otherwise, and ξj (1) = ξ0 (1) − (−1)m for j > 0. So the ξj (6= 0) Pq are pairwise “disjoint”, and ξu = j=0 ξj . One knows that G has just q + 1 orbits on the set U [Aschbacher, 1986, (22.4)]. Hence hξU , ξu iG = hξu2 , 1Gu iG = hπu , 1G i = q + 1. This shows that the ξj are irreducible and pairwise distinct even as characters of G = SUm (q). The final statements in (C3) are due to [Tiep–Zalesskii, 1996]. ¤

Remark . The values of the character ξu and of the irreducible Weil characters ξj on Gu = GUm (q) have been computed in [Gérardin, 1977] and [Tiep–Zalesskii, 1996]. Suppose ξj lies over λju , as before, and let again zu be the generator of Zu = Z(Gu ) with λu (zu ) = ε = e2πi/(q+1) . Let g ∈ Gu , and let dk = dk (g) be the dimension of CU (zu−k g) over Fq2 . Then ξu (g) = (−1)m (−q)d0 and, for j ∈ {0, · · ·, q}, ξj (g) =

q (−1)m X (−q)dk εkj . q+1 k=0

Bibliography Alperin, J. L. (1987), Weights for finite groups, Proc. Symp. Pure Math. 47, 369-379. Aschbacher, M. (1984), On the maximal subgroups of the finite classical groups, Invent. math. 76, 469-514. Aschbacher, M. (1986), Finite Group Theory, Cambridge University Press, Cambridge. Aschbacher, M. (2004), The status of the classification of the finite simple groups, Notices Amer. Math. Soc. 51, 736–740. Bell, G. W. (1978), On the cohomology of the finite special linear groups, I, II, J. Algebra 54, 216-238 and 239-259. Bourbaki, N. (1958), Algèbre, Chap. 8, Hermann, Paris. Brauer, R. (1956), Number theoretical investigations on groups of finite order, Proc. Int. Symp. Alg. Number Theory, Tokyo Nikko, pp. 55–62. Brauer, R. and Feit, W. (1959), On the number of irreducible characters of finite groups in a given block, Proc. Nat. Acad. Sci. U.S.A. 45, 361-365. Brauer, R. (1968), On blocks and sections in finite groups, II, Amer. J. Math. 90, 895-825. Conway, J. H. et al. (1985), Atlas of Finite Groups [Atlas], Clarendon, Oxford. Dade, E. C. (1974), Character values and Clifford extensions for finite groups, Proc. London Math. Soc. (3) 29, 216–236. Dade, E. C. (1992), Counting characters in blocks, I, Invent. math. 109, 187-210. Dickson, L. E. (1901), Linear Groups, Teubner, Leipzig. Dieudonné, J. A. (1971), La Géométrie des Groupes Classique, Springer, Berlin. Dixon, J. D. and Mortimer, B. (1996), Permutation Groups, Springer, Berlin. Erdös, P. (1942), On an elementary proof of some asymptotic formulas in the theory of partitions, Annals Math. (2) 43, 437–450. Feit, W. (1982), The Representation Theory of Finite Groups [F], NorthHolland, Amsterdam. Fong, P. (1962), Solvable groups and modular representation theory, Trans. Amer. Math. Soc. 103, 484–494. 225

226


Gallagher, P. X. (1962), Group characters and commutators, Math. Z. 79, 122–126. Gallagher, P. X. (1970), The number of conjugacy classes in a finite group, Math. Z. 118, 175–179. GAP – Groups, Algorithms, and Programming (2006), version 4.4.9 [GAP], (http://www. gap-systems.org). Gérardin, P. (1977), Weil representations associated to finite fields, J. Algebra 46, 54–101. Gluck, D. (1984), On the k(GV ) problem, J. Algebra 89, 46–55. Gluck, D. and Wolf, T. (1984), Brauer’s height conjecture for p-solvable groups, Trans. Amer. Math. Soc. 282, 137–152. Gluck, D. and Magaard, K. (2002a), The extraspecial case of the k(GV ) problem, Trans. Amer. Math. Soc. 354, 287–333. Gluck, D. and Magaard, K. (2002b), The k(GV ) conjecture for modules in characteristic 31, J. Algebra 250, 252–270. Gluck, D., Magaard, K., Riese, U. and Schmid, P. (2004), The solution of the k(GV )-problem, J. Algebra 279, 694–719. Goodwin, D. (2000), Regular orbits of linear groups with an application to the k(GV )-problem, 1, 2, J. Algebra 227, 395-432 and 433–473. Gow, R. (1980), On the number of characters in a p-block of a p-solvable group, J. Algebra 65, 421–426. Gow, R. (1993), On the number of characters in a block and the k(GV ) problem for self-dual V , J. London Math. Soc. 48, 441–451. Green, J. A. (1955), The characters of the finite general linear groups, Trans. Amer. Math. Soc. 80, 402-447. Griess, R. L. (1973), Automorphisms of extraspecial groups and nonvanishing degree 2 cohomology, Pacific J. Math. 48, 403–422. Guralnick, R. M. and Saxl, J. (2003), Generation of finite almost simple groups by conjugates, J. Algebra 268, 519–571. Guralnick, R. M. and Tiep, P. H. (2005), The non-coprime k(GV ) problem, J. Algebra 293, 185–242. Hall, J. L., Liebeck M. W. and Seitz, G. M. (1992), Generators for finite simple groups, with applications to linear groups, Quart. J. Math. Oxford 43, 441–458. Hering, Ch. (1985), Transitive linear groups and linear groups which contain irreducible subgroups of prime order, II, J. Algebra 93, 151–161. Higman, G. (1960), Enumerating p-groups. I: Inequalities, Proc. London Math. Soc. (3) 10, 24–30. Hilton, P. J. and Stammbach, U. (1971), A Course in Homological Algebra, [HS], Springer, Berlin.

Bibliography

227

Huppert, B. (1967), Endliche Gruppen I, Springer, Berlin. Huppert, B. and Blackburn, N. (1982), Finite Groups III, Springer, Berlin. Isaacs, I. M. (1973), Characters of solvable and symplectic groups, Amer. J. Math. 95, 594–635. Isaacs, I. M. (1976), Character Theory of Finite Groups, [I], Dover, New York. James, G. and Kerber, A. (1981), Representation Theory of the Symmetric Group, Addison–Wesley, London. Jansen, Ch. et al. (1995), An Atlas of Brauer Characters [B-Atlas], Oxford University Press, Oxford. Jennings, S. (1941), The structure of the group ring of a p-group over a modular field, Trans. Amer. Math. Soc. 50, 175–185. Keller, T. M. (2006), Fixed conjugacy classes of normal subgroups and the k(GV )-problem, J. Algebra 305, 457–486. Kleidman, P. and Liebeck, M. W. (1990), The Subgroup Structure of the Finite Classical Groups, Cambridge University Press, Cambridge. Knörr, R. (1984), On the number of characters in a p-block of a p-solvable group, Illinois J. Math. 28, 181–210. Knörr, R. (1990), A remark on Brauer’s k(B)-conjecture, J. Algebra 131, 444–454. ¨ Köhler, C. (1999), Uber das k(GV )-Problem, Diplomarbeit TH Aachen. Köhler, C. and Pahlings, H. (2001), Regular orbits and the k(GV )- problem, in Groups and Computation III (Columbus, Ohio), de Gruyter, Berlin. Kovács, L. G. and Robinson, G. R. (1993), On the number of conjugacy classes of finite groups, J. Algebra 160, 441–460. K¨ ulshammer, B. (1987), A remark on conjectures in modular representation theory, Archiv Math. 49, 396–399. Landazuri V. and Seitz, G. M. (1974), On the minimal degrees of the projective representations of the finite Chevalley groups, J. Algebra 32, 418-443. Liebeck, M. W. (1985), On the orders of maximal subgroups of the finite classical groups, Proc. London Math. Soc. (3) 50, 426–446. Liebeck, M. W. (1996), Regular orbits for linear groups, J. Algebra 184, 1136–1142. Liebeck, M, W. and Pyber, L. (1997), Upper bounds for the number of conjugacy classes of a finite group, J. Algebra 198, 538–562. Maróti, A. (2005), Bounding the number of conjugacy classes of a permutation group, J. Group Theory 8, 237–289. Nagao, H. (1962), On a conjecture of Brauer for p-solvable groups, J. Math. Osaka City Univ. 13, 35–38.

228


Okayama, T. and Wajima, M. (1980), Character correspondences and pblocks of p-solvable groups, Osaka J. Math. 17, 801–806. Rasala, R. (1977), On the minimal degrees of characters of Sn , J. Algebra 45, 132–181. Reynolds, W. F. (1963), Blocks and normal subgroups of finite groups, Nagoya J. Math. 22, 15–32. Riese, U. and Schmid, P. (2000), Self-dual modules and real vectors for solvable groups, J. Algebra 227, 159–171. Riese, U. (2001), The quasisimple case of the k(GV )-conjecture, J. Algebra 235, 45–65. Riese, U. (2002), On the extraspecial case of the k(GV )-conjecture, Archiv Math. 78, 177–183. Riese, U. and Schmid, P. (2003), Real vectors for linear groups and the k(GV )-problem, J. Algebra 267, 725–755. Robinson, G. R. (1995), Some remarks on the k(GV ) problem, J. Algebra 172, 159–166. Robinson, G. R. and Thompson, J. G. (1996), On Brauer’s k(B)-problem, J. Algebra 184, 1143–1160. Robinson, G. R. (1997), Further reductions for the k(GV )-problem, J. Algebra 195, 141–150. Robinson, G. R. (2004), On Brauer’s k(B)-problem for blocks of p-solvable groups with non-Abelian defect groups, J. Algebra 280, 738–742. Sah, C. H. (1977), Cohomology of split group extensions, II, J. Algebra 45, 17–68. Schmid, P. (1981), On the Clifford theory of blocks of characters, J. Algebra 73, 44–55. Schmid, P. (2000), On the automorphism group of extraspecial 2-groups, J. Algebra 234, 492–506. Schmid, P. (2005), Some remarks on the k(GV ) theorem, J. Group Theory 8, 589–604. Schur, I. (1907), Untersuchungen u ¨ber die Darstellungen der endlichen Gruppen durch gebrochene lineare Substitutionen, J. Reine Angew. Math. 132, 85–137. ¨ Schur, I. (1911), Uber die Darstellungen der symmetrischen und alternierenden Gruppen durch gebrochene lineare Substitutionen,J. Reine Angew. Math. 139, 155–250. Seitz, G. M. and Zalesskii, A. E. (1993), On the minimal degrees of projective representations of the finite Chevalley groups, II, J. Algebra 158, 233–234. Serre, J.-P. (2003), On a theorem of Jordan, Bull. Amer. Math. Soc. 40, 429–440.

Bibliography

229

Sin, P. (1996), Modular representations of the Hall–Janko group, Commun. Algebra 24, 4513–4547. Solomon, R. (2001), The classification of the finite simple groups, Bull. Amer. Math. Soc. 38, 315–352. Springer, T. A. and Steinberg, R. (1970), Conjugacy classes, in Seminar on Algebraic Groups and Related Finite Groups, Lecture Notes in Math. 131, pp. 167-266, Springer, Berlin. Steinberg, R. (1967), Lectures on Chevalley Groups, Yale University, Yale. Suzuki, M. (1962), On a class of doubly transitive groups, Annals Math. 75, 105–145. Tiep, P. H. and Zalesskii, A. E. (1996), Minimal characters of finite classical groups, Commun. Algebra 24 (6), 2093–2167. Tiep, P.H. and Zalesskii, A. E. (1997), Some characterizations of the Weil representations of the symplectic and unitary groups, J. Algebra 192, 130–165. Ward, H. N. (1966), On Ree’s series of simple groups, Trans. Amer. Math. Soc. 121, 62–69. Ward, H. N. (1972), Representations of symplectic groups, J. Algebra 20, 182–195. Weil, A. (1964), Sur certain groupes d’opérateurs unitaire, Acta Math. 111, 143–211.

List of Symbols

k(X) = |C`(X)| = |Irr(X)|; (1.1a) Sym2 (V ), Alt2 (V ); 2 Y \X, CoreX (Y ); 3 Y wr Sn ; 3 IndX Y (θ); (1.2b) TenX Y (θ); (1.2e) ωχ ; (1.3a) C`(G|Ω) = orb(G, Ω); (1.4a) πΩ ; 6 Gal(K|Q); 7 (Z/eZ)? unit group; 7 V = CV (G) ⊕ [V, G]; (1.6a) kg (N ) = |CC`(N ) (g)|; 10 IX (θ), Irr(X|θ); 12 µK0 G (θ); 13 M(G) ∼ = H2 (G, C? ); 14 X(θ) = G(θ)∆G X; 15 R = Z(p) [ε]; 19 ωB ; 20 |IBr(X)| = |C`(Xp0 )|; (2.1b) dχϕ , cϕψ ; 21 eB ; 22

J(F X), J(Z(F X)); 25 k(B), `(B); 25 B = bX ; 26 (y,b) dyχϕ , mχζ ; 27 Op (G), Op0 (G), Op0 p (G); 29 V ] = V r {0}; 34 δV = |V |/πV ; (3.3b) H ∗ = Hom(H, C? ); 42 V ∗ = HomF (V, F ); 45 1+2m E∼ , Ω1 (E); 47, 48 = q±,0 C(E) = COut(E) (Z(E)); 48 ε(V ), I(V ), J(V ); 61 P1 (V ); 65 β(G), β ∗ (G); 75 f (g, V ); 75 gV = [d1 , · · ·, dt ]; 76 R0 (L); 116 c(L); 116 µtχ,H ; 117 Y S ; 161 p(n); 170 P (L); 171

Group structures: Zm = m cyclic group of order m, Sm , Am symmetric and alternating group of degree m, respectively, Qm , Dm quaternion and dihedral group of order m, respectively, 1+2m q+ extraspecial group of order q 1+2m and exponent q > 2, 21+2m central product of m copies of Q8 , sign + for even m, ± 21+2m = 21+2m ◦ Z4 , ± 0 X × Y , X ◦ Y direct and central products, respectively, X∆Q Y fibre-product of X, Y over their common quotient group Q, Gwr S wreath product of G with the permutation group S. The notation for the classical groups (groups of Lie type) and sporadic simple groups is standard and follows the [Atlas]. 230

Index

Abelian vector, 44, 63 Absolutely irreducible, 4 Almost quasisimple group, 115 Almost simple group, 115 Alternating square, 2, 215 Atlas, 1 Block, 19 Block induction, 26 Bottom (of a group), 74 Brauer Atlas, 20 Brauer character, 20 Brauer correspondent, 26, 202 Brauer graph, 22 Brauer’s first main theorem, 26 Brauer’s second main theorem, 27 Brauer–Feit theorem, 25 Cartan invariants, 21 Cauchy–Frobenius formula, 6 Central character, 5, 19–31 Centre, centralizer, 1, 2 Class function, 1 Class sum, 1 Clifford–Gallagher formula, 18 Clifford obstruction, 13 Clifford theory, 12–18, 71, 156 Cohomology, 13, 213–216 Commutator, 5, 8, 9 Conjugacy class, 1 Core of a reduced pair, 74 Covering number, 116 Decomposition numbers, 21, 37 Defect, 23 Defect class, 24 Defect group, 24 Deleted permutation module, 63 Dual module, 45

Elementary group, 2 Extended representation group, 16 Extraspecial group, 47–48 Fixed point ratio, 75, 86 Fong–Swan theorem, 31, 210 Frobenius embedding, 3, 149 Frobenius reciprocity, 3 Gallagher, 8, 10, 18 Glauberman, 9 Good conjugacy classes, 10, 18, 54 G-set, 6 Hall–Higman lemma, 29 Height of a character, 23, 202 Higher decomposition numbers, 27, 38 Holomorph, 48–54 Induced block, 26 Induced character, 5 Inertia group, 12 Jacobson radical, 25 Knörr’s generalized character, 37, 196 Large reduced pair, 74 Liebeck–Pyber theorem, 173 Mackey decomposition, 3 Major subsection, 27, 38 Marks of a group, 6, 117 Modular decomposition, 20 Nagao, 31 Nonreal reduced pair, 74, 82, 110 Normal core, 3 n-vectors, 182 231

232


O’Nan–Scott theorem, 172 Orthogonality relations, 1, 2 Orthogonal modules, 45, 60 Parabolic subgroups, 95, 100–109 p-constrained groups, 29, 210 Permutation character, 6, 30 Permutation pair, 63, 112, 154, 199 p-rational character, 31, 54 Principal block, 24 Projective representation, 13 Projective marks, 118 p-solvable groups, 28, 203 Quasisimple groups, 71, 110–147 Quaternion group, 35 Ramification index, 12 Real conjugacy class, group, 7 Real vector, 63 Reduced group, pair, 74 Regular orbit, 8, 63, 112 Regular vector, 32, 63, 112 Representation group, 15 Riese–Schmid theorem, 156

Robinson–Thompson theorem, 66 Root block, 203 Schur cover, 16, 215 Schur index, 2, 5, 16 Schur multiplier, 14, 215 Self-dual module, 44 Spectral pattern, 76 Stabilizer residual, 161 Standard holomorph, 49 Singer cycle, 34, 195 Spectral pattern, 76 Splitting field, 2 Strongly real vector, 63 Subsection, 27, 38 Symmetric square, 2, 215 Symplectic group, 47, 56, 129–136 Symplectic modules, 45, 60 Tensor induction, 4, 70 Trace character, 31, 38 Weil characters, 56, 58, 231–226 Wreath product, 3, 169

The Solution of the k(GV) Problem (ICP Advanced Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

The Geometry of Curvature Homogeneous Pseudo-riemannian Manifolds (ICP Advanced Texts in Mathematics) Imperial College Press - World Scientific

A Solution to the Problem of Induction

XML: Problem - Design - Solution

Problem - Reaction - Solution

Advanced Topics in Computational Number Theory (Graduate Texts in Mathematics)

Advanced Topics in Computional Number Theory (Graduate Texts in Mathematics)

Mathematical Logic: Undergraduate Texts in Mathematics (Undergraduate texts in mathematics)

A Hilbert Space Problem Book (Graduate Texts in Mathematics)

A Hilbert Space Problem Book (Graduate Texts in Mathematics)

Problem-Solving Strategies (Problem Books in Mathematics)

Problem-Solving Strategies (Problem Books in Mathematics)

A Problem Seminar (Problem Books in Mathematics)

The Solution to the 'Son of Man' Problem

Advanced Calculus: A Geometric View (Undergraduate Texts in Mathematics)

Advanced Linear Algebra, Third Edition (Graduate Texts in Mathematics)

The Econometrics of Macroeconomic Modelling (Advanced Texts in Econometrics)

The Geometry of Domains in Space (Birkhäuser Advanced Texts)

The Econometrics of Macroeconomic Modelling (Advanced Texts in Econometrics)

Silverlight 4: Problem – Design – Solution

Numerical Mathematics (Texts in Applied Mathematics)

Algebra (Graduate Texts in Mathematics)

Resemblance nominalism: a solution to the problem of universals

Algebra (Graduate Texts in Mathematics)

Silverlight 4: Problem - Design - Solution

Moduli of Curves (Graduate Texts in Mathematics)

Topology of Surfaces (Undergraduate Texts in Mathematics)

The Solution of the k(GV) Problem (ICP Advanced Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

Advanced Linear Algebra (Graduate Texts in Mathematics)

The Geometry of Curvature Homogeneous Pseudo-riemannian Manifolds (ICP Advanced Texts in Mathematics) Imperial College Press - World Scientific

A Solution to the Problem of Induction

XML: Problem - Design - Solution

Problem - Reaction - Solution

Advanced Topics in Computational Number Theory (Graduate Texts in Mathematics)

Advanced Topics in Computional Number Theory (Graduate Texts in Mathematics)

Mathematical Logic: Undergraduate Texts in Mathematics (Undergraduate texts in mathematics)

A Hilbert Space Problem Book (Graduate Texts in Mathematics)

A Hilbert Space Problem Book (Graduate Texts in Mathematics)

Problem-Solving Strategies (Problem Books in Mathematics)

Problem-Solving Strategies (Problem Books in Mathematics)

A Problem Seminar (Problem Books in Mathematics)

The Solution to the 'Son of Man' Problem

Advanced Calculus: A Geometric View (Undergraduate Texts in Mathematics)

Advanced Linear Algebra, Third Edition (Graduate Texts in Mathematics)

The Econometrics of Macroeconomic Modelling (Advanced Texts in Econometrics)

The Geometry of Domains in Space (Birkhäuser Advanced Texts)

The Econometrics of Macroeconomic Modelling (Advanced Texts in Econometrics)

Silverlight 4: Problem – Design – Solution

Numerical Mathematics (Texts in Applied Mathematics)

Algebra (Graduate Texts in Mathematics)

Resemblance nominalism: a solution to the problem of universals

Algebra (Graduate Texts in Mathematics)

Silverlight 4: Problem - Design - Solution

Moduli of Curves (Graduate Texts in Mathematics)

Topology of Surfaces (Undergraduate Texts in Mathematics)

Recommend Documents